implementation and evaluation of pedagogical strategies in adaptive e-learning ... ·...

Implementation and Evaluation ofPedagogical Strategies in Adaptive

E-Learning Environments

Felix Modritscher

Implementation and Evaluation of Pedagogical Strategies inAdaptive E-Learning Environments

Dissertation in fulfilment of the requirements for the academic degree

Doctor of Technical Sciences (Dr.techn.) in Computer Scienceat the

Graz University of Technology

submitted by

Felix Modritscher

Institute for Information Systems and Computer Media (IICM),Graz University of Technology

A-8010 Graz, Austria

May 2007

c© Copyright 2007 by Felix Modritscher

First reader: Univ.-Prof. Dr.Dr.h.c.mult. Hermann MaurerSecond reader: Univ.-Prof. Dr. Klaus TochtermannAdvisor: Univ.-Ass. Dr. Christian Gutl

Implementierung und Evaluierung padagogischer Strategien inadaptiven Lernumgebungen

Dissertation zur Verleihung des akademischen Grades

Doktor der Technischen Wissenschaftenan der

Technischen Universitat Graz

vorgelegt von

Felix Modritscher

Institut fur Informationssysteme und Computer Medien (IICM),Technische Universitat Graz

A-8010 Graz

Mai 2007

c© Copyright 2007, Felix Modritscher

Diese Arbeit ist in englischer Sprache verfasst.

Erster Begutachter: Univ.-Prof. Dr.Dr.h.c.mult. Hermann MaurerZweiter Begutachter: Univ.-Prof. Dr. Klaus TochtermannBetreuer: Univ.-Ass. Dr. Christian Gutl

Abstract

In our knowledge-oriented society of the 21st century, the necessity for novel teaching and learningparadigms is increasing and, furthermore, is leading to development streams like distance learning or e-learning. However, these new educational approaches often fail due to reasons such as the negligence ofpedagogical principles, a lack of personal contact with the teacher or other learners, usability problems ofthe learning platform, low quality of the learning content and so forth. As one possible answer to suchproblematic aspects, adaptive e-learning deals with implementing typical didactical competencies withininformation technology. Through these methods knowledge transfer can be improved, for example bymeans of observing, assessing and adapting the learning process.

In addressing adaptive behaviour in e-learning environments, this dissertation aims to examine theo-retical and practical aspects of adaptive e-learning and to develop a technological prototype for the AdeLEresearch project. AdeLE, which stands for “Adaptive e-Learning with Eye-Tracking”, focuses on two im-portant requirements. On the one hand, eye-tracking technology ought to be applied for enhanced learnerobservation. Therefore, the prototypical solution should consider the integration of such a device, whilethe usefulness of eye-tracking for adaptive e-learning is decidedly not part of this work. On the other hand,a tool which is named “Dynamic Background Library” and realises the idea of retrieval-based instructionshould be utilised to support the adaptation of the online learning process.

From the theoretical viewpoint, the dissertation first and foremost comprises aspects of adaptation sys-tems and technology-based learning and teaching. A formal specification describing adaptive behaviourin e-learning systems is built up by combining these two scientific fields and surveying historical streamsand systemic types of adaptive educational systems. Continuing with practical issues, this formal modelis applied to derive requirements for standardised, adaptable courseware as well as on an ideal adaptivee-learning environment. Consequently, the technological realisation of the AdeLE prototype is describedin consideration of these requirements. Finally, this work also evaluates the most relevant research ideasof the AdeLE solution approach, namely the prototype itself, the usefulness of the Dynamic BackgroundLibrary and the impact of teacher-driven adaptation of online learning.

Kurzfassung

Der anhaltende Trend zur Wissensgesellschaft schurt den Bedarf nach neuartigen Lehr- und Lernpa-radigmen, was in weiterer Folge zu Entwicklungen wie dem Fernstudium oder e-Learning fuhrt. Aller-dings unterliegen diese neuen Unterrichtsmethoden vielen Problemen, die durch Aspekte wie etwa demVernachlassigen von padagogischen Grundsatzen, dem Fehlen des personlichen Kontakts zum Lehren-den bzw. zu den Mitstudenten, Usability-Schwachen der Lernplattform oder mangelhaften Lerninhaltenbegrundet werden. Als moglichen Ausweg aus diesem Dilemma versuchen adaptive Lernsysteme, didak-tische Kompetenzen – wie das Beobachten, Bewerten und Anpassen des Lernprozesses – technologischabzubilden, um einen effizienteren Wissenstransfer zu erreichen.

Die vorliegende Dissertation zielt darauf ab, theoretische und praktische Aspekte von adaptivem e-Learning zu untersuchen und einen entsprechenden Prototyp fur das Forschungsprojekt AdeLE zu ent-wickeln. Konkret behandelt dieses Projekt, bei dem das Kurzel AdeLE fur “Adaptive e-Learning withEye-Tracking” steht, zwei innovative Ansatze. Einerseits soll ein Eye-Tracking Gerat fur eine erweiterteBeobachtung der Lernenden eingesetzt werden, wobei der zu entwickelnde Prototyp jedoch nur die techni-schen Rahmenbedingungen und nicht den Nutzen dieser Technologie fur adaptives e-Learning adressierensoll. Andererseits ist der Einsatz eines Werkzeugs mit dem Namen “Dynamische Hintergrundbibliothek”zu berucksichtigen, um den Lernprozess durch IR-basierte Instruktionsgewinnung zu adaptieren.

Im Theorieteil dieser Arbeit werden zunachst die Bereiche der Adaptionssysteme bzw. des technologie-basierten Lernens und Lehrens untersucht. Nach Zusammenfuhrung dieser zwei Fachrichtungen und einerLiteraturaufarbeitung der geschichtlichen Entwicklungen bzw. der Systemtypen von adaptiven e-Learningwird eine formale Spezifikation, die das adaptive Verhalten in Lernumgebungen beschreibt, vorgestellt.Dieses formale Modell dient im praktischen Teil der Dissertation zum Herleiten von Anforderungen anstandardisierte, adaptierbare Online Kurse sowie an eine idealtypische adaptive e-Learning Plattform. Inweiterer Folge wird die Entwicklung des AdeLE Systems in Anlehnung an diese Anforderungen beschrie-ben. Schließlich werden die zentralen Ansatze des AdeLE Projekts, also der Prototyp selbst, die Nutzbar-keit der Dynamischen Hintergrundbibliothek fur adaptives e-Learning sowie der Einfluss von didaktischerAdaption des Lernprozesses auf den Wissenstransfer, evaluiert.

I hereby certify that the work presented in this dissertation is my own and that work performed byothers is appropriately cited.

Ich versichere hiermit, diese Dissertation selbstandig verfaßt, andere als die angegebenen Quellenund Hilfsmittel nicht benutzt und mich auch sonst keiner unerlaubten Hilfsmittel bedient zu haben.

xi

So long, and Thanks for All the Help!

This dissertation was made possible with the help of many people involved in my professional andprivate surrounding. To those the following words of appreciation are addressed.

First of all, I have to thank all my colleagues at the Institute for Information Systems and ComputerMedia at the Graz University of Technology, not only for enabling and supporting my research work, butalso for managing a very comfortable and productive working environment. Particularly, I am indebtedto my supervisor, Hermann Maurer, for giving me inspiration and valuable feedback. Many thanks go toKlaus Tochtermann being second reader of this dissertation.

Additionally, I would like to offer my gratitude towards all members of the Web Application Group,especially Christian Gutl for recruiting me for the AdeLE project and facilitating my research work. Cor-dial thanks go to Victor Manuel Garcıa-Barrios with whom I did not only share the office, but also manyvaluable ideas and the travail of writing a dissertation. It was his extraordinary way of scientific thinkingthat had a strong impact on my own research activities. Furthermore, I have to thank the project leadersHelmut Leitner and Walter Schinnerl as well as all other members of our group for their excellent team-work and their high-quality output, which was the key factor for the successful completion of so manyprojects in the last six years.

In context of my research work, I have to acknowledge the collaboration with our partners. Represen-tatively for all the others, I thank Maja Pivec and Jurgen Pripfl from the Department of Information Designof the University of Applied Sciences JOANNEUM, Dietrich Albert from the Cognitive Science Sectionof the University of Graz, Alexandra Sindler from the University of Graz, the people from the KnowCenter with whom I was engaged during the APOSDLE project, Heinz Dreher from Curtin University ofTechnology and Elizabeth Peterson from the University of Auckland.

Besides, I am also indebted to the University of Applied Sciences Campus02, where I have heldlectures and supervised 17 diploma theses in the last 4 years. Namely, I have to thank Alfred Zindes, FranzPucher, Valentin Gillich and Georg Lindsberger. Besides lecturing, being lectured in a special course ofstudies on didactics and pedagogy at Campus02 has had positive influence on my personal developmentand my research work. Particularly, I would like to thank my diploma students for their participation andinterest and their excellent theses as well as the 38 students of the class IT02 who participated in a casestudy for my dissertation.

Last but not least, I have to articulate words of thankfulness to those of my private surrounding,precisely my mother Monika, my brother Klaus, his wife Manuela, and their child Melissa. But above andbeyond, I am deeply indebted to my companion in life, Silke Spiel, for her love and patience, for puttingup with late nights, and for even proof-reading and reviewing parts of this dissertation. In the end, it wasprimarily her credits that the quality of my English has significantly improved and that this dissertationcame to a successful end.

In beloved memory of my father Wilhelm

Felix ModritscherGraz, Austria, May 2007

xiii

Contents

1 Introduction 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Objectives of this work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Methodology and structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.4 Scientific contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

I. Theoretical Background 7

2 Adaptation Systems 92.1 Roots and related fields of systems theory . . . . . . . . . . . . . . . . . . . . . . . . . . 92.2 Further developments in systems theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.3 Towards adaptation and related concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.4 A generic approach to adaptation systems . . . . . . . . . . . . . . . . . . . . . . . . . . 202.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Technology-Based Learning and Teaching 273.1 Relevant learning theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273.2 E-pedagogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.3 E-didactics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353.4 Towards formalising e-learning and e-teaching . . . . . . . . . . . . . . . . . . . . . . . . 403.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4 Adaptive E-Learning 474.1 Historical streams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474.2 Types of adaptive educational systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514.3 Existing theoretical models for adaptive e-learning . . . . . . . . . . . . . . . . . . . . . 544.4 Formalising adaptive behaviour in e-learning systems . . . . . . . . . . . . . . . . . . . . 594.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

II. Practical Aspects 63

5 Towards Standardising Adaptable Courseware 655.1 Standardisation of learning content . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655.2 Requirements for standards to support adaptive e-learning . . . . . . . . . . . . . . . . . . 685.3 Inspection of current specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715.4 A standard-based approach to adaptive e-learning . . . . . . . . . . . . . . . . . . . . . . 735.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

xv

6 An Ideal Environment for Adaptive E-Learning 796.1 Methods and techniques for adapting the learning process . . . . . . . . . . . . . . . . . . 79

6.2 Functional requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

6.3 Architectural design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

6.4 Inspecting existing projects and solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 87

6.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

7 Technical Realisation of the AdeLE System 937.1 Planning of the AdeLE prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

7.2 Functional units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

7.3 Implementation details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

7.4 A walk through the AdeLE system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

7.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

III. Proof of Concept 121

8 Adaptation of the Learning Process within the AdeLE Prototype 1238.1 Planning stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

8.2 Experiences gained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

8.3 Other results from literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

8.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

9 Utilising a Dynamic Background Library for Adaptive E-Learning 1339.1 Basic concept and realisation of EHELP . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

9.2 Evaluating the EHELP system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

9.3 The Dynamic Background Library for the AdeLE prototype . . . . . . . . . . . . . . . . 139

9.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

10 The Impact of a Didactical Strategy on Learning 14310.1 Realisation of the courses regarding the learning theories . . . . . . . . . . . . . . . . . . 143

10.2 Comparison of the three e-learning strategies . . . . . . . . . . . . . . . . . . . . . . . . 145

10.3 Findings on didactical and pedagogical aspects . . . . . . . . . . . . . . . . . . . . . . . 150

10.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

11 Conclusions and Outlook 15711.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

11.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

11.3 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

Bibliography 161

Index 181

xvi

List of Figures

1.1 Overview of and connections between the nine chapters of this work . . . . . . . . . . . . 3

2.1 Formal description of a generic system . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.2 Spectrum of adaptation in computer systems, adopted from [Oppermann, 1994] . . . . . . 162.3 Generic framework for an adaptation system . . . . . . . . . . . . . . . . . . . . . . . . 212.4 Formal specification of a multi-purpose adaptive system (types and variables) . . . . . . . 222.5 Formal specification of a multi-purpose adaptive system (operations and thread) . . . . . 23

3.1 Overview of research issues related to the learning process . . . . . . . . . . . . . . . . . 413.2 Formal specification of the content model . . . . . . . . . . . . . . . . . . . . . . . . . . 423.3 Formal specification of the pedagogical model . . . . . . . . . . . . . . . . . . . . . . . 433.4 Formal specification of the didactical model (types and instance variables) . . . . . . . . 443.5 Formal specification of the didactical model (operations) . . . . . . . . . . . . . . . . . . 45

4.1 Model of adaptive instruction, adopted from [Park et al., 1987] . . . . . . . . . . . . . . . 554.2 Framework for adaptive e-learning, adopted from [Shute and Towle, 2003] . . . . . . . . 564.3 The KnowledgeTree architecture, adopted from [Brusilovsky, 2004b] . . . . . . . . . . . 574.4 Formal specification of the adaptation model . . . . . . . . . . . . . . . . . . . . . . . . 61

5.1 The development process of e-learning standards, adopted from [Gries, 2003] . . . . . . . 675.2 Enhancing SCORM’s structuring and content packaging specification . . . . . . . . . . . 745.3 Enhancing SCORM’s asset specification . . . . . . . . . . . . . . . . . . . . . . . . . . . 755.4 “Semantic TAGging Editor” for inner-instructional objectives . . . . . . . . . . . . . . . 76

6.1 Architectural design of an adaptive e-learning environment . . . . . . . . . . . . . . . . . 85

7.1 Utilisation of the Tobii 1750 Eye-Tracking system . . . . . . . . . . . . . . . . . . . . . 957.2 Overview of AdeLE’s architectural design . . . . . . . . . . . . . . . . . . . . . . . . . . 967.3 Implementation details of the Adaptive System . . . . . . . . . . . . . . . . . . . . . . . 987.4 Implementation details of the Modelling System, adapted from [Froschl, 2005, p. 118] . . 1007.5 Graphical user interface of the Modelling System [Froschl, 2005, p. 154] . . . . . . . . . 1017.6 Implementation details of the Concept-Based Context Modeller, adapted from [Safran,

2006, p. 84] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1027.7 Graphical user interface of the Concept-Based Context Modeller [Safran, 2006, p. 106] . . 1037.8 Top of the Openwings Explorer displaying installed components . . . . . . . . . . . . . . 104

xvii

7.9 Bottom of the Openwings Explorer displaying interfaces and services . . . . . . . . . . . 105

7.10 Sequence diagram for the scenario “learner navigates instruction” . . . . . . . . . . . . . 107

7.11 AdeLE system login dialog . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

7.12 Registration form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

7.13 AdeLE prototype main menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

7.14 Dialog for course enrolment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

7.15 Overview of the learning progress for an example course . . . . . . . . . . . . . . . . . . 112

7.16 Form-based dialog to edit the user profile . . . . . . . . . . . . . . . . . . . . . . . . . . 112

7.17 Learner’s view of a course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

7.18 Example for an examination including an assignment task . . . . . . . . . . . . . . . . . 114

7.19 Tree-view navigation of the AdeLE system . . . . . . . . . . . . . . . . . . . . . . . . . 114

7.20 View of the navigation area with hidden elements . . . . . . . . . . . . . . . . . . . . . . 115

7.21 “Background Knowledge” section for an exemplary instruction . . . . . . . . . . . . . . . 115

7.22 “Why this way?” section for an example learner . . . . . . . . . . . . . . . . . . . . . . . 116

7.23 Form-based eye-tracking simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

7.24 Menu with additional functions for teachers . . . . . . . . . . . . . . . . . . . . . . . . . 117

8.1 Distribution of WAVI-factors given by the VICS tool (yellow triangles), students’ self-assessment (green diamonds) and the AdeLE system after completing the course (redsquares) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

9.1 Basic functionality scheme of a DBL [Garcia-Barrios et al., 2002] . . . . . . . . . . . . . 135

9.2 EHELP viewing mode “embedded hyperlinks” [Garcia-Barrios et al., 2002] . . . . . . . . 136

9.3 Background knowledge data structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

10.1 Distribution of teacher’s (blue), students’ (pink) and overall (yellow) activities course A . 147

10.2 Distribution of teacher’s (blue), students’ (pink) and overall (yellow) activities course B . 148

10.3 Distribution of teacher’s (blue), students’ (pink) and overall (yellow) activities course C . 149

10.4 Comparison of the students’ activities for the courses A (green), B (yellow) and C (blue) . 154

xviii

List of Tables

3.1 Bloom taxonomy [Bloom, 1956], adapted and extended for skills and attitudes . . . . . . 37

8.1 Characteristics of the students’ learning behaviour for each initial WAVI-group . . . . . . 127

8.2 Characteristics of the students’ learning behaviour for each pass . . . . . . . . . . . . . . 128

10.1 Statistics of the course’s educational objectives . . . . . . . . . . . . . . . . . . . . . . . 144

10.2 Characteristics of the three courses for the preparation stage . . . . . . . . . . . . . . . . 146

10.3 Characteristics of the three courses for the implementation stage . . . . . . . . . . . . . . 146

10.4 Characteristics of the three courses for the concluding stage . . . . . . . . . . . . . . . . 150

10.5 Objectives, competencies according to the Bloom taxonomy (Type: knowledge, skill, orattitude; Level: 1-6) and the rates of attempts and successful achievements (overall andfor each course) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

10.6 Results of the post-questionnaires on the learner characteristics (each statement rated witha number between one and five comprising the range from “absolute disagreement” to“strong agreement”) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

10.7 Characteristics of the three courses based on the students’ ongoing documentation aboutlearning and raw database queries within the Moodle system . . . . . . . . . . . . . . . . 154

10.8 Results of the post-questionnaire concerning the factors relevant to learning (each state-ment rated with a number between one and five comprising the range from “absolutedisagreement” to “strong agreement”) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

xix

Chapter 1

Introduction

“ A master can tell you what he expects of you.A teacher, though, awakens your own expectations. ”

[ Patricia Neal ]

Experienced teachers tend to possess the ability to awake the learners’ interests and keep them mo-tivated in dealing with the content of a course. In addressing recent streams of educational trends, suchas distance learning, these approaches lack such pedagogical competencies in the learning phase and,therefore, often fail in practice.

1.1 Motivation

The shift from a production-centred to a knowledge-centred society, as outlined for example by [Probstet al., 2000, p. 1], had a deep impact on education in general and cognitive science in particular. As knowl-edge turns into a valuable asset for both companies and humans [Maurer, 1998], research and developmentfocus on new educational methods. Technology-based learning and teaching can mainly be identified asone of the emerging areas, as shown by means of concrete numbers in [Brennan, 2003] or [Hasebrookand Maurer, 2004, p. 7ff]. Furthermore, [Dietinger, 2003, p. 21f] points out several application scenarioswhich manifest the necessity of these new approaches of teaching and learning for educational institutionsas well as for companies.

Nevertheless, e-learning initiatives often fail due to problematic aspects, like high costs or politicalinfluence [Noble, 2001], the focusing on technology and the negligence of pedagogical principles [Parket al., 1987], usability problems of e-learning systems [Ardito et al., 2004], etc. To overcome such prob-lems, ideas and solutions are being developed from two directions: On the one hand, cognitive psychol-ogy is heading towards new pedagogical approaches, like shifting to constructivistic learning [Lennonand Maurer, 2003], promoting collaboration and communication [Hooper and Hannafin, 1991], Web 2.0approaches [Kolbitsch and Maurer, 2006] and so forth. On the other hand, technologists are bringing up in-novative concepts and methods, such as game-based learning [Prensky, 2001], virtual campus approacheslike ViKar [Kuhn and Gudjonsdottir, 1999] or automatic adaptation of the online learning process.

Against this background, many scientific and commercial activities address this research and devel-opment stream called “adaptive e-learning” [Shute and Towle, 2003]. [Brusilovsky, 2004a] states thatadaptive educational hypermedia (as one part of adaptive e-learning) is relatively new and started around1990. Nevertheless, as a result of inconsistent definitions of terms as well as missing links between techni-cal approaches and pedagogical basics, the history of commonly-known principles and technical solutions

1

2 1. Introduction

can be traced back to the early 20th century [Pressey, 1926]. According to [Corno and Snow, 1986], sometheoretical ideas date back as far as the 4th century BC.

In a similar way to e-learning, historically two streams can be identified in the field of adaptive e-learning: First, psychologists and educators promoted didactic-driven approaches like the macro-adaptiveinstructional design and systemic types such as computer-managed instruction [Park and Lee, 2004].These concepts and solutions have lead to areas like adaptive educational hypermedia and have had animpact on learning content management and e-learning standards. Second, technologists have aimed ateven more algorithmic directions and developed the micro-adaptive instructional approach and technolo-gies like intelligent tutoring systems based on the fundaments of systems theory.

All in all, [Brusilovsky and Peylo, 2003] state that in the last decade a new generation of adaptiveeducational systems has been established by combining both the content-based and the algorithmic ap-proaches mentioned above. Therefore, adaptive e-learning environments are capable of (automatically)assessing learner characteristics and states and, further, adapting the learning content as well as featuresof the platform towards changes in connection with the learning process. As this behaviour is a prerequi-site skill for teaching in the classroom, adaptive e-learning can be understood as one possible attempt toimplement pedagogical competencies within information technology.

1.2 Objectives of this work

Tying up to the need for the adaptation of the online learning process, this dissertation aims at the devel-opment of a framework for adaptive e-learning and its applicability in practice. Therefore, the theoreticalbasics of adaptive e-learning ought to be derived from two directions: On the one hand, systems theoryshould give an overview of relevant terms and concepts of adaptation systems in general. On the otherhand, technology-based learning and teaching address pedagogical and didactical issues for realising on-line courses and distance learning phases.

As a result of these two theoretical streams and on the basis of a literature survey on the history andtechnologies of adaptive e-learning, a theoretical framework for the use of formally describing adaptivebehaviour in e-learning environments is developed. By utilising this framework, practical aspects suchas providing adaptable courseware and functional requirements for adaptive e-learning systems ought tobe examined. Additionally, this work describes the development of a technological framework and aprototype for the research project AdeLE [AdeLE, 2006].

AdeLE is the abbreviation of “Adaptive e-Learning with Eye-Tracking” and stands for a researchproject carried out from 2003 to 2007. Specifically, AdeLE aims at the development of a technology-based solution exploiting novel methods for retrieval-based instructions and fine-grained user profilingbased on real-time eye-tracking and content-tracking information. As this dissertation primarily focuseson adaptive e-learning environments, the usefulness of eye-tracking techniques is decidedly not part ofthis work.

Finally, this work will also evaluate different aspects of adaptive e-learning: Firstly, the theoreticalframework is evaluated by its application in the practical part of this dissertation. Secondly, the usefulnessand usability of the AdeLE prototype has to be examined. Thirdly, the applicability of the “DynamicBackground Library”, a tool to retrieve instructions dynamically from other repositories, is shown in thecontext of this work. Fourthly, the overall idea of adapting the online learning process is evaluated bymeans of examining the impact of didactical (teacher-driven) adaptation of learning.

1.3. Methodology and structure 3

1.3 Methodology and structure

In order to achieve these objectives, the dissertation is separated in three main parts, as visualised in figure1.1:

• After this introductory chapter, the first part consists of three chapters on the most relevant theoret-ical areas for this work, namely adaptation systems, technological-based learning and teaching andadaptive e-learning.

• The second part describes practical aspects of adaptive e-learning, comprising aspects of standard-ised learning content, general requirements for an adaptive e-learning environment as well as thetechnical realisation of the AdeLE prototype.

• Finally, the third part deals with the proof-of-concept for adaptive e-learning. Thus, the usefulnessand usability of the AdeLE prototype, the applicability of the Dynamic Background Library foradapting the learning process and the impact of didactical strategies on online learning are evaluated.

Figure 1.1: Overview of and connections between the nine chapters of this work

Starting with the theoretical part of the dissertation, chapter 2 examines the history and relevant re-search streams of adaptation systems on the basis of an extensive literature survey and introduces the

4 1. Introduction

most relevant concepts in this area. Further, a general model for adaptation systems is developed. Dueto the fact that e-learning comprises human-system-interaction, the literature survey and the theoreticalframework include, amongst others, aspects of systems thinking, problem situations or human systems.

Chapter 3 deals with technology-based learning and teaching. Therefore, the three relevant learningtheories, i.e. Behaviourism, Cognitivism and Constructivism, are briefly highlighted. Further, pedagogicalaspects such as those factors which influence learning or the learner characteristics are pointed out. Afterdiscussing didactical issues like defining competencies and learning objectives for a course or determin-ing appropriate teaching and assessment methods, this chapter concludes with an overview of importantconcepts and a formal model of technology-based learning and teaching.

In chapter 4, the central core of the dissertation, a generic framework for adaptive e-learning, isintroduced. After reviewing historical approaches and adaptive educational systems, existing theoreticalmodels are introduced and inspected on the basis of findings on adaptation systems and technology-basedlearning and teaching. Finally, prior to the conclusion of the theoretical part of this work, a new attempttowards formalising adaptive e-learning is presented.

After these theoretical chapters, the practical part of this work starts with chapter 5 which deals withstandardised adaptable courseware. Thus, the standardisation process within the scope of e-learningis outlined and requirements for standards to support adaptive e-learning are manifested. Subsequently,the current specifications and standards are inspected and a selection of specifications enabling adaptivee-learning is established for the AdeLE project.

Thereafter, chapter 6 highlights commonly-known adaptation methods and techniques and derivesconcrete requirements for an ideal adaptive e-learning environment from the generic framework of thetheoretical part. Further, an architectural design for this system is presented and existing projects as wellas commercial products are examined according to the requirements stated in this chapter.

In chapter 7 the realisation of the AdeLE prototype is summarised from the planning stage to adetailed description of the system. In addition to outlining the special requirements given by the eye-tracking device and the Dynamic Background Library applied in this project, all functional units of thesystem are pointed out. Further, relevant implementation details are explained, and the adaptation modelis described using a walkthrough of the prototype.

After pointing out the technical realisation of the AdeLE system, the third part of the dissertationattempts to prove the concept of “adaptive e-learning”. Therefore, chapter 8 starts with an evaluation ofthe adaptation of the learning process within the AdeLE prototype. After describing the setup of anevaluation study, experiences gained through this study are summarised and improvements of the AdeLEprototype are presented. Further, on the basis of a literature survey, benefits of various adaptive e-learningtechniques and solutions are highlighted.

In chapter 9, the usefulness of a Dynamic Background Library is evaluated within the scope ofadaptive e-learning. Hereby, the basic concept of a dynamic background library, which comprises theidea of retrieval-based instruction, is introduced and the applicability of the prototypical implementationis examined by a case study. The chapter then describes experiences gained by this study in order toredesign and realise the idea of a Dynamic Background Library for the AdeLE system.

Finally, chapter 10 describes another case study on the comparison of three different e-teachingmethods. After pointing out the strategy of the case study and showing the implementation of the differentcourses, the implementation of the e-learning phase is depicted in detail and experiences about the needto adapt teaching methods are highlighted. Concluding the study and the proof-of-concept for adaptivee-learning, the dissertation is summed up and an outlook for further work is presented.

1.4. Scientific contributions 5

1.4 Scientific contributions

This dissertation covers the results of four years of research work (2003-2007) in the field of e-learning.This work primarily deals with the AdeLE approach (“Adaptive e-Learning with Eye-Tracking”, [AdeLE,2006]) and, thus, treats theoretical and practical issues of adaptive e-learning. Nevertheless, the authorhas also participated in other e-learning projects, namely in the “Learning Northern Ireland” (LNI)project [C2k, 2003] as well as in the EU project “Advanced Process-Oriented Self-Directed LearningEnvironment” [APOSDLE, 2007]. Overall, the impact in connection with the aforementioned projects onthis work is limited, as they do not include a variety of aspects of adaptive e-learning.

In the context of these projects, the following parts of this dissertation have already been published invarious conference proceedings and journals:

• Major parts of chapter 3 (technology-based learning and teaching) are based on [Modritscher,2006a], [Modritscher, 2006b], [Modritscher and Sindler, 2005] and [Modritscher et al., 2006b].

• Chapter 4 (adaptive e-learning) includes sections of [Modritscher et al., 2004b] and [Modritscher,2006c].

• In chapter 5 (standardisation of adaptable courseware) parts of [Modritscher et al., 2004c] and[Modritscher et al., 2004a] were reused.

• Chapter 7 (the description of the AdeLE prototype) is primarily based on [Modritscher et al., 2006a].A few paragraphs were also taken from [Gutl and Modritscher, 2005], [Gutl et al., 2004], [Gutl et al.,2005] and an internal project report.

• Unpublished material about an evaluation study has been used for chapter 8 (evaluation report onthe first AdeLE prototype).

• Chapter 9 (the usefulness of a Dynamic Background Library for adaptive e-learning) consists of afew parts of [Modritscher et al., 2005] and [Garcia-Barrios et al., 2004a], but mainly summarisesthe corresponding evaluation study.

• Finally, chapter 10 (a study comparing different e-teaching strategies) is based on [Modritscher,2006a], [Modritscher, 2006b] and [Modritscher et al., 2006b].

Overall, this work aimed at contributing to different theoretical and practical issues within the scopeof adaptive e-learning. Thus, the following research results have to be emphasised at this point:

• From a theoretical viewpoint, the first three chapters give an overview of basic concepts of adapta-tion systems, technology-based learning and teaching and adaptive e-learning. Particularly, existingmodels for adaptive e-learning were examined with regard to the relevance and applicability forthe AdeLE approach. As none of them covers the full spectrum of the theoretical issues of thiswork, informal frameworks of an adaptation system and of technology-based learning and teachingwere developed by the author of this work. Further, a novel model to formally describe adaptivebehaviour in e-learning environments (FORMABLE) was built up by utilising the specification lan-guage VDM++.

• With respect to further practical issues, requirements for adaptable courseware were derived fromthe FORMABLE model. Consequently, existing standards and specifications were evaluated ac-cording to these requirements and a standard-based approach was built up for the AdeLE system.In this context, the author identified various shortcomings of existing standards and specifications,which was approved by Peter Brusilovsky a few months after these findings had been published.

6 1. Introduction

• Moreover, an ideal environment for adaptive e-learning was developed in order to examine variousresearch prototypes and products. Considering adaptive methods and techniques given by literature,the FORMABLE model was applied to determine requirements and an architectural design for anadaptive e-learning system and, further, to inspect selected research projects and solutions in orderto identify adaptive features.

• Thereafter, the AdeLE project and its special requirements were introduced, whereby particularlythe development process and the resulting prototype were outlined. The learning platform and theAdaptive System of the AdeLE prototype were developed by the author. Moreover, the integrationof the modelling systems and the eye-tracking device and the implementation of client-sided toolswere also results of the practical part of this dissertation. Overall, the two special requirements ofthe AdeLE approach can be considered to be fulfilled. On the one hand, the technical feasibility ofintegrating eye-tracking technology was shown and, additionally, the usefulness was confirmed bymeans of learner state modelling within the AdeLE prototype. On the other hand, the idea of theDynamic Background Library was considered and realised within the system.

• To evaluate the benefits of adaptive e-learning, three assumptions were examined in terms of casestudies: (1) the usefulness and usability of the AdeLE prototype, (2) the usefulness of adaptinginstructions through retrieval-based mechanisms and (3) the impact of an e-teaching strategy ononline learning. The study on the AdeLE system was planned and exploited by the author, while thecase study of the impact of e-teaching strategies on online learning was done both in the lecture andseparately. Concerning the evaluation of the Dynamic Background Library, the author supportedthe evaluation team in running the study and exploiting the results.

• Basically, the proof-of-concept section of the dissertation lead to three concrete results: Firstly, theadaptation model of the AdeLE prototype is rather prototypical and, from viewpoint of cognitivescience, insufficient. Further, the first version of AdeLE also lacked important usability issues.Secondly, the original concept of the Dynamic Background Library might not be satisfactory for theAdeLE system. Thus, a redesign towards retrieval-based instruction was conducted. This approachis very similar to the basic idea of the technological outcome of the APOSDLE project, which hasbeen started about half a year after the implementation of the Context Modeller. Thirdly, the casestudy of the influences of e-teaching strategies on the learning process manifested that didactic-driven adaptation is useful or even necessary for online courses.

I. Theoretical Background

7

Chapter 2

Adaptation Systems

“ To improve is to change;to be perfect is to change often. ”

[ Winston Churchill ]

Adaptation is definitely not a new concept, in particular when it comes to considering related termsderived from a great variety of different research areas such as biology, climatology, cybernetics, infor-mation theory, system theory and so forth. Therefore, a system might be able to adapt its behaviour insome way to fulfil a certain purpose, for instance to compensate for environmental changes [Ashby, 1960,p. 207], to accommodate changes in the dynamics, parameters or disturbances of processes [Ogunnaikeand Ray, 1994, p. 1087], etc.

As a basis for further treatment with adaptive e-learning, this chapter deals with systems adapting partsof themselves in general. Therefore, section 2.1 discusses some ideas from systems theory, philosophyand methodology. Further, section 2.2 points out the relevant developments of systems theory. Section2.3 defines the basic terms which are necessary in the context of adaptation systems and examines theseconcepts and systemic characteristics by means of a theoretical model. Finally, section 2.4 introduces aformal approach for an adaptation system which is based on the theoretical findings of this chapter.

2.1 Roots and related fields of systems theory

According to [Banathy and Jenlink, 2004], systems inquiry consists of three interrelated research areas:systems theory, systems philosophy and systems methodology. While systems theory is about devel-oping concepts and principles concerning systems in various sciences, systems philosophy deals with thereorientation of thought and worldview. Systems methodology comprises models, strategies and toolsapplying systems theory and philosophy in practice according to the analysis, the design, the developmentand the management of complex systems.

9

10 2. Adaptation Systems

2.1.1 History of systems theory

Tracking back the roots of system theory, in the middle of the 20th century some basic principles andconcepts of a general system theory were set up by researchers coming from different sciences. Yet, thehistory of systems theory has started much earlier as the following review shows:

• Some of the key concepts of systems theory – such as composing various interacting parts intoa whole, the separation from its environment and so forth – were firstly defined by the Germanphilosopher Georg Wilhelm Friedrich Hegel in the 18th century.

• [Francois, 1999] reports that the French psychologist Claude Bernard who had worked on theinternal milieu of living beings from 1854 to 1878 stated that there is a “difference between theprocedures inside a system and its environment”. Further, in 1866, the de Cyon brothers discoveredthe first example of a self-regulating system, “the countervailing action of the accelerator and themoderator nerves of the heart”.

• According to [Banathy and Jenlink, 2004], the term “general theory of systems” was first used bythe Hungarian philosopher and scientist Bela Zalai who developed his theory in the years 1913 and1914. In addition, in his research work carried out from 1921 to 1927, the Russian scientist Alexan-der Alexandrovich Bogdanov dealt with the field of Tektology, “a dynamic science of complexwholes concerned with universal structural regularities, general types of systems, the general lawsof their transformation and the basic laws of organisation” [Bogdanov, 1984, p. ii (by Gorelik)].

• In 1932, the French scientist Walter Cannon introduced the biological concept of homeostasis,which can be interpreted as an extension of Bernard’s idea of stability in the internal milieu. Be-ginning with 1942, the French biologist Pierre Vendryes studied the feature of regulation in livingand non-living systems, extensively developing the concept of autonomy.

• The fields of Cybernetics and System Theory, two disciplines with a large history and often seen asthe roots of fields like Artificial Intelligence or Control Systems, were represented in the 1940’s bya melting pot of research groups, where ideas matured, a general set of vocabularies for engineeringand physiology were introduced and the basic terms of system theory – such as learning, regulation,adaptation, self-organisation, perception, memory and so forth – were created. Referring [Wiener,1948], the American mathematician Norbert Wiener founded the discipline of Cybernetics in 1948.

• In the 1950s, the well-known Austrian-born biologist Ludwig von Bertalanffy introduced the ideaof a General System Theory (GST) redefining the boundaries of other disciplines such as mathe-matics, biology, biophysics, psychology etc. [VonBertalanffy, 1956] defines systems as “complexesof elements standing in interaction” and deals with concepts like organisation, wholeness, direc-tiveness, teleology, control, self-regulation, differentiation and so forth. As a consequence, thisapproach justifies the existence of general systems properties as well as organisation of complexity.

• In 1954, along with the American economist Kenneth Boulding, the American neurophysiologistRalph Gerard and the Russian-born mathematician-biologist Anatol Rapoport, von Bertalanffyfounded the Society for General System Theory – later renamed to International Society for theSystems Sciences (ISSS). Without using the term GST, a similar theory was developed by the Britishpsychiatrist William Ross Ashby during the years 1945 and 1947, as published for example in[Ashby, 1960].

Overall, systems theory is an important movement to develop and evaluate systems with respect toa certain behaviour given by requirements, relevant properties and interaction with its environment. Asshown later on, adaptation systems require various concepts which are outlined in the (incomplete) histor-ical overview above.

2.2. Further developments in systems theory 11

2.1.2 Systems philosophy

Dealing with aspects of systems theory’s benefits, [Banathy and Jenlink, 2004] state that “systems philos-ophy is concerned with a systems view of the world and systems thinking as an approach to theoreticaland real-world problems”. Therefore, philosophical aspects are worked out in three directions: Firstly,the ontological approach describes the systems view of the world. Secondly, epistemology researches thesystem-internal representation of the world’s view. And finally, axiology is directed to the study of value,ethics and aesthetics guided by the moral question about its meaning.

While the axiological concern of systems philosophy aims to ensure that systems inquiry is moraland ethical and that the participants in system inquiry, i.e. developers, are constantly questioning im-plications of their actions, ontological and epistemological aspects deal with researching and realisingsystem-internal models about the real world.

2.1.3 Systems methodology

Systems methodology can be considered as an important part of systems inquiry and, referring to [Banathyand Jenlink, 2004], holds two important domains: “the study of methods by which we generate knowledgeabout systems in general” as well as “the identification and description of strategies, models, methodsand tools for the application of systems theory”. Thus, systems methodology can be seen as a set ofmethods and tools relevant to three application areas: (1) the analysis of systems and problems concernedwith systemic aspects, (2) the design, development, implementation and evaluation of systems and (3) themanagement of systems and changes in systems.

Using systems methodology, four steps have to be regarded: First of all, the problem situation hasto be identified, characterised and classified. Secondly, the problem context must be determined anddescribed. Thirdly, the type of system in which a certain problem situation is embedded has to be locatedand specified. And finally, methods for specific strategies, tools for the given problem situations as wellas the type and the context of a system must be selected. Nowadays, the appropriate methodology canbe chosen from a wide range of methods and tools that are available. Exemplarily, the RESCUE processintroduced by [Jones and Maiden, 2005] can be mentioned here as a methodology to develop and evaluatesocio-technical systems such as e-learning environments.

Summarising this section, it has to be stated that systems philosophy, systems theory and systemsmethodology are applied in the functional context of systems. Systems philosophy comprises methodsfor defining and organising the principles constituting systems theory. Moreover, systems theory pro-vides guidance for selecting, developing and organising approaches, methods and tools given by systemsmethodology. As new application areas – for example approaches in the field of adaptive e-learning –continue to make ever greater demands on systems and increase the complexity of them, researchers haveto consider systems theory and other research fields dealing with systems.

2.2 Further developments in systems theory

This section introduces selected research fields which were originated on the basis of the General SystemTheory and are relevant to this work. With respect to adaptation systems and, further, to adaptive e-learning, these six main areas are examined in the following subsections: (1) Hard-systems science, (2)cybernetics, (3) systems thinking, (4) human systems, (5) problem situations and (6) systems design.


2.2.1 Hard-systems science and cybernetics

Referring to [Banathy and Jenlink, 2004], hard-systems science comprises two mainstream develop-ments: Firstly, operations research developed during the 1950s as a result of the complexity of logisticsand resource management in the Second World War aims at a quantitative analysis of rather closed sys-tems. Secondly, systems engineering “is concerned with the design of closed man-machine systems andlarger scale sociotechnical systems”, inter alia by applying specific methods and tools, activities to solveproblems and a set of relations between the tools and the activities.

[Kossiakoff and Sweet, 2003, p. 50ff] define the systems engineering process as a life cycle consistingof three stages (with three phases each): (1) concept development (with the need for analysis, conceptexploration and concept definition), (2) engineering development (advanced engineering, engineering de-sign, integration and evaluation) and (3) post development (production, operation and support). Systemsengineering can be seen as an essential research field for many practical areas, such as software develop-ment, project management and so forth.

Cybernetics deals with the self-organisation of human, artificial and natural systems. This researchfield was developed through two phases, the first-order cybernetics focusing on observed systems and thesecond-order cybernetics also including the observing system. [Wiener, 1948] introduced cybernetics asthe science of “communication and control in the animal and the machine” comprising the processes ofself-organisation and self-regulation by means of the concept of first-order cybernetics. These principlesmay be applied on both computers and humans (cognitive sciences) and may be seen as the heart ofneural network approaches in computing. Moreover, [VonFoerster, 1984, p. 288ff] treats second-ordercybernetics which supplies the need of a “theory of the observer” (contrary to observed systems whichcan be understood as first-order cybernetics).

Both hard-systems science and cybernetics are of enormous importance for adaptation systems. Whilethe systems engineering process aims at developing and evaluating the (technological) solution itself,cybernetics helps describe and understand the behaviour of such a system. Adaptive behaviour is not onlyrelated to self-organisation and self-regulation (first-order cybernetics), but also to observing a part of thesystem (second-order cybernetics).

2.2.2 Systems thinking

The evolution of systems thinking resulted in the fact that traditional paradigms based on analytic think-ing, reductionism and determinism are insufficient (e.g. for areas like artificial intelligence or adaptivesystems). Hence, the need for “intelligent” behaviour of systems was articulated. According to [Banathyand Jenlink, 2004], the most important mainstream developments can be itemised as follows: LivingSystems Theory, General Theory of Dynamic Systems, Unbounded Systems Thinking, Critical SystemsTheory, Liberating Systems Theory and the application of post-modern theory to systems theory.

In the scope of this work, the Living Systems Theory is of particular importance. “The philosophy oforganism” manifested in [Miller and Miller, 1995] extends the organismic orientation of von Bertalanffy’sGeneral System Theory and describes eight levels of living systems: cell, organ, organism, group, organi-sation, community, society and supernatural system. The central thesis of this theory is that a system canbe characterised by 20 subsystems – components for processing information, matter, or energy – whoseprocesses are essential to life. In fact, information systems and cognitive sciences primarily deal with theconcepts of information processing (virtually on each of the eight levels of living systems).

2.2.3 Human systems

Human systems focus both on systems in general and on their application on social or human systems.Therefore, [Checkland, 1981, p. 14ff] introduces so-called human activity systems (HAS), which comprise

2.2. Further developments in systems theory 13

the perceptions of human actors using a system and structure groups of people working with a system onthe basis of their activities. Checkland defined a human activity system as a “notional system that expressessome purposeful human activity that could be found in the real world”.

[Banathy, 1988] reports on a classification scheme for HAS on the basis of four dimensions: (1) theopenness of human systems, (2) their mechanistic vs. systemic nature, (3) the unitary vs. pluralisticdefinition of their purpose and (4) the degree and nature of their complexity. As a result, the followingtypes of systems can be identified:

• Rigidly Controlled Systems are rather closed systems, have a simple structure and consist of onlya few elements with limited interaction between them. Normally, they fulfil one single purpose, likeassembly-line or man-machine systems, for example.

• Deterministic Systems are systems which can be defined as more closed than open and have clearlydefined goals. Their complexity ranges from simplicity to detail. Within such a system, a user hasa limited degree of freedom in the choice of the methods. Examples of deterministic systems arebureaucracies or instructional systems.

• Purposive Systems, on the other hand, are defined as more open than closed, but are unitary. Theyreact to environmental influences in order to maintain their viability and allow people to selectoperational means and methods completely freely, but with respect to the system’s purpose. Corpo-rations, social service agencies or public education systems can be stated as examples of purposivesystems.

• Heuristic Systems formulate their own goals with respect to policy guidelines. They are open tochanges and even initiate such changes. Thus, they are based on a dynamic complexity, while theirinternal arrangements and operations are systemic, as exemplified with innovative business venturesor alternative educational systems for example.

• Purpose-Seeking Systems seek their own ideals and are guided by their own rules and vision. Beingconsidered as open, they evolve while co-operating with their environment. They are characterisedby a dynamic complexity and systemic behaviour as they constantly seek new purposes and searchfor new niches in their environment. Examples are cutting-edge R&D agencies or communitiesseeking to integrate their development with other organisations.

In the context of this work, classification of human systems is useful to assess and describe the extentby which a system can be adapted or can adapt itself. An information system in general can be consideredto be deterministic, but an adaptation system might be classified as a purposive, heuristic or even as apurpose-seeking system. As a conclusion, openness, complexity and purposes are important issues foradaptation systems.

Addressing methodological aspects of HAS, the development of such systems requires socio-technicalapproaches like RESCUE [Jones and Maiden, 2005] or, within the scope of e-learning in Small-to-MediumSized Enterprises (SMEs), the E-Learning Ecosystem (ELES, [Chang and Gutl, 2007]) which comprisesan ecological, application-based approach to design e-learning environments. Overall, developing HASesis a rather complex process due to the nature and complexity of human-system-interaction.

2.2.4 Problem situations and systems design

Referring to human systems, the area of problem situations which deals with systems comprising similarproblems can be identified. Although problems are unbounded and may be regarded to be subjective,[Ackoff, 1981, p. 79ff] suggests that “a set of interdependent problems constitutes a system of problem”,


namely a mess. A mess has, as any other system, properties that none of its parts has and, additionally, islost when the system is taken away or when the system is separately considered.

[Ulrich, 1983, p. 308ff] points out the difference between ill-defined (or ill-structured) problems andwell-defined problems in which the initial constitutions, the goals and the necessary operations are spec-ified. As a consequence, science should only deal with well-structured or well-defined problems. Nev-ertheless, many problems – in particular social problems – are naturally non-deterministic. Thus, everysolution to these problems can be considered to be incomplete. Ill-defined problems require continuoussystems design until the problem situation is clearly defined.

With the consideration of a design problem to be ill-structured, [Checkland, 1981, p. 149ff] statesthat hard systems thinking fails, if systems design lacks the naming and definition of its objectives, whilehuman systems deal with valuing of individuals, collectives and the role of technology. Further, thedesign approach focuses on several various segments of the system, in particular on problem situationsand solutions. [Churchman, 1971, p. 4] outlines that, if decision rules affect the state of the whole system,the designer should not separate parts of it merely for stability reasons. Overall, systems design can becomprised as a hard process including a large amount of communication. Successful design implies thatknowledge can be transferred into action, into another design for example.

Furthermore, [Checkland, 1981, p. 163ff] describes the development of a Soft Systems Methodology(SSM) for the work with human activity systems and the use of system ideas to define basic mental pro-cesses: (1) perceiving, (2) predicating, (3) comparing and (4) deciding to take action. This methodologyis derived from the concept of human activity systems by means of the attributes which are important forall human activities. To summarise systems design, it has to be stated that designers have to consider anentity to be designed as a whole in terms of the synthesis of the interaction of its parts. Therefore, all partsof the system need to be designed interactively and at the same time, which requires co-ordination anddesign for interdependency across all system levels that need to be integrated.

Generally, this section outlined six important mainstream developments founded on the basis of sys-tems theory and relevant to adaptation systems. While hard-systems science deals with the design of thesystem itself, cybernetics, systems thinking and human systems can be used to characterise and understandadaptation processes within a system, for example by defining its complexity, examining information pro-cessing processes and units and so forth. Finally, an analysis of problem situations as well as the aspectsof systems design that apply might be necessary to avoid ill-defined systems and to guarantee importantsystem properties such as the stability, the usefulness, etc. Although there are much more research fieldsrelevant to adaptation systems – for example human-computer-interaction, software development, etc.,these areas are omitted in this work, because they are not necessary for the theoretical model described inthe next two sections.

2.3 Towards adaptation and related concepts

After summing up important concepts of systems theory and related mainstream developments, this sec-tion defines and explains the most important terms in connection with adaptation systems to provide abasic vocabulary for the other chapters of this dissertation. Following systems methodology, the formalspecification language VDM++ (see [Fitzgerald and Larsen, 1998]) is applied to build up a theoreticalmodel of a general system and examine aspects of adaptation within this system.

2.3. Towards adaptation and related concepts 15

According to [Kossiakoff and Sweet, 2003, p. 3], a system can be defined as “a set of interrelatedcomponents working together toward some common objective”. This definition implies that a set of inter-acting parts collectively perform a significant function. Generally, a system can formally be described as acollection (map) of components. Further, a component consists of a unique entity (within the system) andowns a certain state (see figure 2.1). Although this specification appears to be rather simple to examine allaspects of systems theory – it does not allow overlapping components for example – it suffices to describeadaptation systems as demonstrated in this section.

Figure 2.1: Formal description of a generic system

Based on the formal description of a generic system the following subsections deal with basic con-cepts, components, models and system properties relevant to adaptation systems. Moreover, the conceptof adaptation is examined in connection with users interacting with the system.

2.3.1 Adaptation, adaptability and adaptivity

[DeJong, 1975, p. 5] defines adaptation as “a strategy for generating better-performing solutions to aproblem by reducing the initial uncertainty about the environment via feedback information made avail-able during the evaluation of particular solutions”. In the context of systems, the concept “adaptation”describes the process of modifying a system in some way to reach a certain goal. According to the formalmodel of a generic system, adaptation within a system can be defined as changing the state of at least oneof the system’s components. As a result, the overall system, or parts of it, could behave in a different wayand, further, system properties might change as well.

In this context, adaptation systems comprise all systems providing the possibility that the state of atleast one component can be modified. Exemplary systems allowing the adaptation of internal states canbe found in a broad range of fields, reaching from microcontroller over computer systems to organisationsin any kind of application area. A state-based approach to systems theory, like the Turing Machine or theVon Neumann Machine described in [Mills, 2006, p. 17ff], may be of relevance for adaptive e-learningenvironments, as will be outlined later.


[Oppermann, 1994] introduces a “spectrum of adaptation in computer systems” and differentiatesbetween adaptivity (adaptive) and adaptability (adaptable). As both concepts deal with adaptationsystems, adaptability comprises the idea that the user initiates the adaptation, while adaptivity is aboutsystems automatically adapting themselves according to some prescription. As shown in figure 2.2, vari-ous steps between full adaptive behaviour and pure adaptability can be identified.

Figure 2.2: Spectrum of adaptation in computer systems, adopted from [Oppermann, 1994]

With respect to the formal description of a generic system, adaptability addresses the adaptation trig-gered by a user, while adaptivity implies that at least one component is able to modify the state of at leastanother component of the system. According to [Brusilovsky, 2003a], meta-adaptation implies that auser or the system itself adapts the components which are responsible for automatically modifying statesof other components.

2.3.2 Necessary components and models

The concepts of adaptation, adaptability and, particularly, adaptivity require a number of special com-ponents within a system. Concerning adaptation in general, the set of adaptable objects comprises allcomponents which the state can be altered for. Thus, an adaptation system must have at least one of thesecomponents.

To realise real adaptive behaviour within a system, two important systemic parts are required: Onthe one side, an adaptive component, which can be defined as a component adapting the state of othercomponents, is necessary. In addition, an adaptive engine (or “adaptive system”) comprises all adaptivecomponents within a system. On the other side, a system-driven adaptation of components requires variousmodels to decide, which components of the system have to be adapted on the basis of which information,when, in which way and why (see also [Brusilovsky, 1996] or [Specht, 1997, p. 14ff]):

• First of all, adaptation information determines what is used as a source of adaptation. Users orthe knowledge about users are often considered to be relevant to adaptation, as can be concludedfor example by [Benyon, 1993]. However, adaptation can be based on any state in the system’senvironment. Therefore, an adaptation system includes at least one component observing its envi-ronment and assessing the relevant states. The set of these components defines the environmentalmodel, which might, for instance, be a user model.

• Secondly, adaptation rules are necessary for the decision to start the adaptation process. Theserules based on the adaptation information can be regarded as the trigger for adaptation. With re-spect to the formal description of a generic system, this rule engine could be realised as an owncomponent.


• Thirdly, the adaptation procedures comprise which systemic components are adapted in whichway. Thus, certain components – so-called adaptors – are responsible to modify the adaptableobjects according to their internal prescriptions.

• Finally, adaptation targets might be related to adaptation information but aim to formalise thegoals of adaptation (and not its source). Thus, components modelling adaptation targets also haveto observe and assess environmental states and, in addition, are not used as a basis for adaptation,but to evaluate its effects in the real world. Generally, adaptation targets can be considered asinferences on the usefulness of a system’s adaptive behaviour and, therefore, have to be evaluatedby the system designer.

As mentioned above, adaptation systems require at least the first three models – adaptation informa-tion, rules and procedures – to provide real adaptive behaviour. These three models are subsumed as theadaptation model of an adaptive system. Furthermore, the definition and assessment of adaptation tar-gets is necessary to evaluate the value of the system-driven adaptation in practice. From the viewpoint ofadaptability, it can be outlined that the user determines the adaptation information, the adaptation rulesand the adaptation procedures by changing the states of the adaptable objects. As a conclusion, these threemodels are redundant for systems providing only adaptable features.

In the formal description of an adaptive component, it is possible to subsume all components necessaryfor system-driven adaptation within one component. On a higher level of living systems theory, this partof the adaptation system can be considered as an adaptive engine. Although a great variety of these mono-lithic solutions exist in this area, current research approaches focus on distributing adaptation systemsby means of their main components like an adaptive system or modelling system (e.g. see [Brusilovsky,2004b], [Conlan, 2005, p. 100ff] or [Modritscher et al., 2006a]).

Finally, a meta-adaptation system is characterised by the fact that at least one adaptive componentis also part of the adaptable objects. Thus, the behaviour of this component can be adapted by a useror another “meta-adaptive” component and, in consequence, the adaptive behaviour of the whole systemmight also be subject to change.

2.3.3 Examination of systemic characteristics

In order to examine characteristics of adaptation systems, the following concepts of cybernetics, humansystems and problem situations (see section 2.2) are applied:

• Firstly, an adaptation system implementing adaptive features can be considered to act as an observer,so that it can be classified within the scope of second-order cybernetics. On the other hand, systemsproviding adaptable functions can be seen as observed systems by means of first-order cybernetics.

• Secondly, the application of Benathy’s classification scheme for human activity systems on adapta-tion systems leads to the following conclusions: Systems providing adaptation in some way mightnot be rigidly controlled, but they ought to be at least deterministic, by reaching their adaptationtargets for example. Generally, adaptation systems behave like purposive systems trying to react toenvironmental influences. At a sophisticated level, it is also possible that an adaptation system mayformulate its own goals according to some policy. Yet, real purpose-seeking behaviour by meansof adjusting the adaptation targets can be considered as hardly realisable within a system lackinghuman intelligence.

• Thirdly, addressing problem situations in the field of adaptation systems, it is recommended that theadaptation process itself must be specified as a well-defined and well-structured problem. In otherwords, the adaptation targets – but not the system itself – must be deterministic in order to provide


valuable adaptation. An ill-defined adaptation process might lead to non-deterministic situations,such as confusing the user, worsening the systems usability or even cancelling its usefulness, asstated by [DeBra, 2000].

As a result of these considerations, systemic characteristics which are relevant to adaptation systemsare examined according to the formal model of a generic system and defined as follows:

• As mentioned above, the complexity of a system is determined by the internal structure, for instanceaccording to von Bertalanffy’s GST, as well as by its behaviour, with respect to cybernetics forexample. Other properties like the system’s flexibility, its reliability or its overall behaviour highlydepend on this internal structure and the interaction between the components.

• Another characteristic relevant to adaptation systems is the concept of self-organisation which de-scribes a system’s ability to reorganise its components and their interactions. As a result, a systemmight provide another or even a new structure or systemic behaviour, as stated for example in [Ger-shenson and Heylighen, 2003]. Concerning adaptation, self-organising systems can be determinedas adaptive systems (see also [Heylighen, 2003]).

• According to [Skyttner, 2001, p. 58], openness defines how a system is dependent upon its environ-ment, to exchange matters, energy and information for example. Thus, an open system allows itselfto be controlled by external information or fuelled from outside by some form of energy, while aclosed system does not provide interfaces to its environment. Moreover, autonomy in this contextmeans that a system acts on its own, without being dependent on other environmental entities.

• As part of a system’s openness, observability comprises the concept that the system’s behaviouris observable from the outside in some way. To realise an observable system, it needs to be opento enable insights into its functionality. Yet, this information channel into the system might notprovide the possibility to control the system. As explained in the next subsection, observability isof particular importance to user-adaptive systems.

• Then again, controllability also deals with openness, but requires an information channel control-ling the system. Again, this systemic property is relevant to adaptable systems, i.e. to allow a usermodifying a component’s state, as well as for adaptive behaviour to let the user the control theadaptation process.

• Purposiveness sums up all purposes a system serves. Although this term is rather a philosophicalone – as stated for example by [Baz 2005] – it can be used to characterise a system according to itspurposes. Such a classification scheme can reach from a single-purpose over a multi-purpose up to apurpose-seeking nature. In the context of adaptivity, meta-adaptivity can be seen as purpose-seekingbehaviour.

• Intelligence and learnability describe two further closely related terms. As these two characteris-tics are often associated with humans, intelligent or learnable systems require that adaptation targetsare considered in the future systemic behaviour. Similarly to the idea of a purpose-seeking nature,learnable systems such as Intelligent Tutoring Systems (ITS) are hard to be realised and often workin a very restricted domain or context only, as stated for example by [Park and Lee, 2004].

• The concept of feedback addresses two important issues of adaptation systems. On the one side,feedback deals with controllability of systems which is particularly necessary for adaptable systems.On the other side, feedback also comprises the assessment of states of a system’s environment. Asalready mentioned, feedback is needed to guarantee accurate and timely models of the adaptationinformation and, if available, the adaptation targets.


• In the other direction, feedforward describes the adjustment of a system’s output according to anideal model. In certain situations adaptive systems could also implement some kind of feedforwardstrategy.

In addition to these systemic attributes, there are a great variety of other concepts and characteristics,in particular about evolutionary, structural or behavioural aspects of systems (e.g. stability, vulnerability,sensitivity, etc.). As these aspects are not principally relevant to this dissertation, they are not discussed anyfurther. However, in practice, adaptation systems often deal with user-centred issues, which are examinedin the following subsection.

2.3.4 Adaptation towards users

Many approaches and solutions within the scope of adaptation systems aim to adapt to users, i.e. oftendescribed with concepts like personalisation, customisation, user-adaptive systems or even adaptive sys-tems. In addition, [Weibelzahl, 2003, p. 18f] itemises typical functions of adaptive systems – for examplesupplying to find information, tailoring information, recommending digital artefacts, adapting the user in-terface, etc. [Dreher et al., 2004a] – each one dealing with the user as the adaptation target and adaptationinformation at the same time.

Moreover, some researchers consider personalisation to be a synonym for adaptivity, as outlined bythe following exemplary statements: [Benyon and Murray, 1993] define an adaptive system as a system“which can alter aspects of their structure, functionality or interface in order to accommodate the differingneeds of individuals or groups of users and the changing needs of users over time”. [Jameson, 2001, p. 4]considers a user-adaptive system to be an “interactive system which adapts its behavior to each individualuser on the basis of nontrivial inferences from information about that user”.

Contrary to these viewpoints, the formal model of a generic system (see figure 2.1) and the definitionsof former subsections allow a precise definition of concepts related to user-based adaptation:

• As mentioned above, adaptability means that the user modifies states of a system’s components.Concerning the adaptation target, this kind of adaptation might have an effect on general aspects,for example the system’s internal structure, data persistence processes, etc., or on aspects concern-ing the user, such as the system’s usefulness or usability. The last case comprises the concept ofcustomisation, which can be seen as a user-driven adaptation in order to accommodate needs ofone self or of other users.

• Similarly, personalisation describes the process of automatically adapting systemic states towardsthe needs of a user. Contrary to adaptivity, personalisation implies that the user is the adaptationtarget, while adaptive behaviour might aim at any other issue which is not relevant and visible forthe user of the system. The personalisation process requires a user model which has to be appliedas adaptation information and might also be utilised to model the adaptation targets for evaluationreasons.

In the context of user-centred adaptation, two important characteristics can be outlined as follows:Scrutability, as mentioned and recommended by [Kay, 2000], comprises the extent of systemic observ-ability necessary for a user to understand the personalisation process which is based on an internal usermodel. On the other side, controllability – as defined in the last subsection – has to include aspects ofpersonalisation, so that the user has sufficient control over the adaptation process. However, controllabilityof a system itself could be an adaptation target as well. At least, the degree of controllability ought to bedetermined carefully, as shown in a study by [Jameson and Schwarzkopf, 2002].


To characterise user-centred adaptation, [Garcia-Barrios et al., 2005] identified the following five di-mensions of personalisation:

• Explicit vs. implicit personalisation: Explicit personalisation describes the adaptation accordingto a concrete user model. For instance, a system could provide tailored information by using aspecific user profile. Implicit personalisation is about adaptation resulting from a certain context(situation or environment) and without utilising user information explicitly, like adapting the userinterface according to the output device just used by the user for example.

• Perceivable vs. hidden personalisation: Personalisation is called perceivable, if a user recognisesthe results of personalisation. For instance, a system may show or hide control elements, such asa tree-view visualising the structure of the information space according to user-specific cognitivetraits. Hidden personalisation does not affect the user interface or the presented content at all.Taking the following example: If it is intended to update a user model according to some adaptationrule, the user may not recognise the result of this personalisation step.

• Predictive vs. deterministic personalisation: Predictive personalisation comprises adaptationsteps executed in advance. For instance, rearranging the sequence of information chunks to be pre-sented next is not only hidden, but also predictive. Deterministic personalisation takes place withinone adaptation step, if chosen information chunks are aggregated to one page which is immediatelydisplayed for example.

• Controlled vs. uncontrolled personalisation: Controlled personalisation enables and enhancesthe system’s scrutability and, further, describes the idea that the user may take control of adaptationprocesses at any time. Considering the simple case that a system cannot decide the adaptation step,the user could be asked to determine the adaptation, by providing a set of suggestions for example.Uncontrolled personalisation would not allow the user to influence the adaptation process.

• Individual vs. stereotyped personalisation: Individual personalisation comprises personalisationtowards one specific person, while stereotyped personalisation deals with personalisation towardsgroups or anonymous users.

Considering these dimensions of personalisation might be useful for planning and realising person-alisation features. Overall, this section examined the basics of adaptation systems and defined relevantconcepts and terms as consequently used in this dissertation. Furthermore, a generic framework for adap-tation system being based on these fundamentals is introduced in the following section.

2.4 A generic approach to adaptation systems

As a conclusion from the theoretical part about adaptation systems, the idea of a multi-purpose adaptivecomponent and the feasibility of its realisation are discussed in this section. In addition, it is explained howvarious concepts of systems theory and adaptation systems are considered within this approach. Finally,technological aspects and open issues of the multi-purpose adaptive component are addressed before thischapter is concluded.

2.4.1 A framework for an adaptation system

Combining the theoretical concepts of the last section into one piece may lead to a framework as shownin figure 2.3. In general, an adaptation system can be understood as system providing the possibility thatsome internal states can be modified in one of the following two ways: On the one side, a user could

2.4. A generic approach to adaptation systems 21

simply adjust these adaptable objects manually. On the other side, an adaptive component could observedifferent models and initiate the adaptation process on the basis of rules and by means of executing so-called adaptors.

Figure 2.3: Generic framework for an adaptation system

The left side of figure 2.3 shows the adaptive behaviour which consists of different models, a triggerand a set of adaptors implementing the adaptation itself. The models can be seen as internal representationsof the real world and might be located within the adaptive component, but also be provided by othercomponents, by an own modelling component or even by an external system for example. The modelshave to be updated according to states from the real world, by environmental or user states or even statesof systemic components for example. Applying a user model as adaptation information would lead topersonalisation, as already stated in the last section.

The adaptive component consists of the internal models, a trigger applying the adaptation rules andthe adaptor implementing the adaptation procedures. Further, the adaptors are responsible for carrying outthe adaptation process by modifying the systemic states of the adaptable objects. As a matter of course,an adaptor might also be a part of the set of adaptable objects, which would describe the concept of meta-adaptivity. Moreover, adaptation information or adaptation targets can also focus on systemic states, toadapt according to some systemic aspect or to measure the adaptation effects for example.

The right side of the framework shown in figure 2.3 describes the process of adaptability and cus-tomisation. If a user is allowed to modify systemic states without having a benefit, the system can beconsidered to be adaptable. Additionally, if user-triggered adaptation of systemic states also changes theusefulness or usability, a system is also customisable.

2.4.2 Formal model of a multi-purpose adaptive system

In the context of the generic framework for adaptation systems, the adaptive component is of particu-lar relevance. While adaptability and customisation can be realised easily, adaptivity and personalisa-tion requires an adaptive component, which is more complex, because adaptive behaviour requires morefunctions as outlined in the last section. Considering purposiveness, this subsection describes a genericframework for a multi-purposive adaptive system by utilising the formal specification language VDM++as addressed for example in [Fitzgerald and Larsen, 1998].


In accordance with the formal model of a generic system given in the last section, a generic adaptivesystem as defined in this subsection can be described as collection of components consisting of an entityand a state. A model as well as an adaptor is specified equivalent to a component (see also figure 2.1),whereby the model comprises a real component within the adaptive system, while the adaptor only de-termines which component has to be set to which state. Thus, an adaptor is more a special kind of statecomprising a prescription for adaptation.

Figure 2.4: Formal specification of a multi-purpose adaptive system (types and variables)

The section instance variables in figure 2.4 demonstrates that an adaptive system is equivalent to ageneric system, but includes models and adaptors. As already mentioned, a model is simply a systemiccomponent with an own state, while an adaptor can be seen as a component with a state defining theadaptation effect. The invariants in this model guarantee that all entities – even the ones of the adaptors’adaptation prescriptions – are part of the adaptive system. Although it would be possible to deal withexternal components as well, this formal model is restricted to systemic components to prohibit unnec-essary complexity. Nevertheless, this model allows external components to be addressed as shown in thefollowing.

2.4. A generic approach to adaptation systems 23

Figure 2.5 sums up the operations and the main thread of the generic adaptive system. Referring tothe generic framework for an adaptation system in the last subsection, three operations are of importance:Firstly, assessing the adaptation information comprises the process of accurately mapping environmentaland user states to the appropriate adaptation procedures. Secondly, observing the models deals with eval-uating whether one of the adaptation rules is triggered. Thirdly, performing the adaptation specifies theapplication of the adaptation procedures by means of executing the adaptors.

Figure 2.5: Formal specification of a multi-purpose adaptive system (operations and thread)

Although the first two operations “assess” and “observe”, require a sequence of models, there is oneimportant difference. The assessment of the adaptation information requires information about externalstates and the components to store them. Observation addresses the internal models and their states totrigger the adaptation process. Further, the “adapt”-operation processes a list of adaptors to manipulatesystemic states according to the adaptation procedures or user-given input. As it is not specified if theoperations are triggered by a human or another system, it is possible to consider components of externalsystems, observing external models or adapting states in external systems for example.

Without any further concept, this formal model would comprise a system providing adaptability, i.e.a user triggers the updates of the internal models as well as the adaptation. The thread-section in figure2.5 extends this formal approach towards adaptive behaviour by means of continuously evaluating theadaptation rules and adapting systemic states if the observation of internal models is triggered. Therefore,the operations “observer” and “adapt”, which are manifested in literature for example by [Brusilovsky,1996] or [DeKoch, 2000, p. 15], are applied.

Additionally, the assessment of the internal models could also be performed pro-actively, but as this isnot the main task of an adaptive component, it is left out here. Moreover, the assessment of external states


might be independent of the adaptation process itself and be part of some own modelling component orthe user. As a result, this mechanism updating the internal models according to real-world states has to besupplied by another trigger, by another systemic component, an external system or a user for example.

Concluding this formal approach towards a multi-purpose adaptive system, it has to be stated thatadaptation targets can be seen as part of the systemic models and, further, meta-adaptation is realisableby adapting models or adaptors. Therefore, the evaluation and (automatic) improvement of the adaptationprocess itself is realisable within this framework. Further, aspects of systemic intelligence and learnabilitywould imply a set of rules implementing meta-adaptive behaviour. Thus, aspects of Artificial Intelligence(AI) are also considered by this formal model.

2.4.3 Practical application and exemplary systems

In practice the framework for an adaptation system as well as the formal model for a multi-purpose adap-tive system might be utilised to design and realise adaptive or adaptable behaviour within informationsystems. Moreover, this approach can be also used to evaluate existing solutions towards adaptationmechanisms. Finally, the idea of developing a multi-purpose adaptive component can be pointed out as abenefit, as many projects and solutions throughout all possible research fields deal with adaptation systemsor, at least, require functions of adaptability or adaptivity.

From the technological viewpoint the following observations concerning the technical realisation ofadaptation systems have to be manifested here:

• Firstly, the increasing complexity of applications and systems lead to technological concepts, suchas distributed systemic architectures or service-oriented approaches. The formal model of ageneric system (see figure 2.1) as well as the one of a multi-purpose adaptive system (see figure2.4 and figure 2.5) support distributed environments, i.e. with the component-based description ofa system. These models also consider different abstraction levels, reaching from an architecturaloverview of a system to a detailed specification of modules or services. Because these formalspecifications provide a lot of freedom for distributed architecture, systems engineers have to ensurethat the number of distributed entities is not too large, as shown for example by a case study in [Gutland Garcia-Barrios, 2005b].

• Secondly, an adaptive component might require some data layer, i.e. a database managementsystem, for different reasons. On the one hand, a data layer is necessary to store and manageadaptation rules. On the other hand, adaptation decisions have to be logged for scrutability reasons,as mentioned in the last section.

• Thirdly, the adaptation described by the formal model in this section is based on a rule-based ap-proach. Nevertheless, adaptive behaviour of systems often deals with artificial intelligence, forexample in the area of control systems as outlined by [Tsypkin, 1971] or, if human expertise shouldbe simulated, as stated by [Park and Lee, 2004]. However, artificial intelligence methodology isconsidered by the formal model of a multi-purpose adaptive system in two ways: On the one side,an AI-method can be formalised by a set of rules, as given for instance in expert systems [Gi-arratano and Riley, 1994]. On the other side, more sophisticated AI-functionality, like a neuralnetwork [Abdi, 1994], can be described with a meta-adaptive system, in which the internal models,the adaptors or even the adaptation is in control.

2.5. Conclusions 25

When applying the framework for adaptation systems, in practice the following types of systems canbe evaluated and characterised with the approach in this section:

• On the lowest level of adaptation, systems without any feature of adaptivity or adaptability can bedescribed. In that case, the formal model would be reduced to the one of a generic system givenin figure 2.1. While the actions “observe” and “adapt” as well as adaptation rules and adaptors areredundant, assessment of real-world states might be necessary. A typical example of such a systemwould be a user profiler which only tracks information about the user without deriving or providingany models [Gutl and Garcia-Barrios, 2005b].

• On the next evolutionary step the formal approach is applicable for all systems providing someadaptability. Thus, the assessment of environmental and user states as well as the adaptation istriggered by the user. Generally, all systems providing some kind of user settings – for example forcustomisation reasons – can be assigned to this systemic type.

• Further, adaptation systems with simple and static adaptation rules have to be mentioned here. Forinstance, a system enabling such behaviour might provide automatic adaptation of a visual elementaccording to the brightness of the environment’s backlight.

• The main group of systems of interest to the formal model of this section comprises so-called single-purpose solutions, which is a valid term for many adaptation systems. Examples are, amongst a largeset of existing systems, A-MEDIAS (an adaptive event notification service [Hinze, 2003, p. 129ff]),solutions in the scope of adaptive hypermedia [Brusilovsky, 1996] and various system types in thefield of adaptive e-learning as shown in the upcoming chapters of this work.

• As already mentioned in this subsection, AI-methodologies can be described with the formal modelof a multi-purpose adaptive system, whether by masquerading the AI-method with adaptation rulesor by defining meta-adaptive behaviour.

• Finally, aspects of user-centred adaptation have to be outlined here. On the one side, customisationcan be described as realised on the basis of systemic adaptability, i.e. by allowing users to customisea system according to their preferences. On the other side, automatic adaptation towards a user leadsto the commonly-known application area of personalisation and recommendation systems, as oftendealt with in fields like knowledge management or technology-based learning (see also [Hicks andTochtermann, 2001] or [Pivec and Baumann, 2004]).

Overall, the generic approach to adaptation systems in this section is very suitable to evaluate or planany kind of adaptable or adaptive feature within a system. As a consequence, this theoretical frameworkis going to be applied in the field of adaptive e-learning in chapter 4.

2.5 Conclusions

Adaptation systems are, as outlined in this section, strongly related to and a part of systems theory. Inthe context of sciences like philosophy, psychology, biology and so forth, researchers started to deal withaspects of systemic organisation, behaviour and characteristics more than 100 years ago, which resultedin different areas, such as hard-systems science, cybernetics, systems thinking, human systems or systemsdesign. Additionally, philosophical and practical issues arose in the scope of systems theory.

Against this background, aspects of adaptation systems can be identified in all of these research fields,beginning with systemic characteristics like complexity, self-organisation, openness, observability, con-trollability, etc. over philosophical aspects such as internal models of the real-world up to systems method-ology for realising such systems. In accordance with these findings and based on a formal model of a


generic system, the most relevant concepts of adaptation systems – adaptation, adaptability and adaptiv-ity – were defined and elements necessary for the adaptation process were described. Moreover, relatedconcepts like meta-adaptivity and user-centred adaptation were also examined.

Based on the historical review of systems theory and the definition of basic concepts in the scope ofadaptation systems, this section attempted to build up a formal framework for a multi-purpose adaptivesystem, which can be applicable for planning or evaluating adaptation systems. As adaptation systems arenot the main issue in this dissertation, the formal approach is not evaluated any further, but utilised forexamining theoretical and practical issues in the scope of this work. Therefore, the next section gives anoverview of technology-based learning and teaching to provide important basics for the central part of thiswork, which deals with adaptive e-learning in theory and practice.

Chapter 3

Technology-Based Learning and Teach-ing

“ To lecture, or to be lectured: that is the question. ”

[ Shakespeare’s Hamlet, freely adapted by the author ]

E-learning – as a synonym for technology-based learning and teaching – is identified as one of theemerging areas, as shown by means of concrete numbers in [Brennan, 2003] and has turned out to beimportant for educational institutions as well as for companies, as highlighted by concrete applicationscenarios in [Dietinger, 2003, p. 21f]. Nevertheless, various problematic aspects such as higher costs andpolitical influence [Noble, 2001], the focusing on technology and the negligence of pedagogical principles[Park et al., 1987], usability problems of e-learning systems [Ardito et al., 2004], etc. were reported.According to [Gunawardena and McIsaac, 2004], a shift from technology to pedagogy-based research canbe observed within the field of distance learning. Educators have become more interested in examiningpedagogical themes and strategies within online courses instead of experimenting with new technologies.

[Jain et al., 2002, p. xi] states that e-learning concerns learning as well as teaching. Therefore, thischapter contributes to the application of technology for distance education along three dimensions. Firstof all, section 3.1 examines generally accepted learning theories in the area of distance learning. There-after, section 3.2 comprises pedagogical issues, like factors influencing the learning process or learnercharacteristics and section 3.3 deals with didactical aspects of online courses, such as the definition andassessment of learning objectives. Concluding this chapter, section 3.4 introduces a framework giving anoverview of relevant concepts and a formal model technology-based learning and teaching.

3.1 Relevant learning theories

With respect to [Oblinger and Hawkins, 2005], “e-learning” is currently used for different educationalscenarios in literature. Therefore, at this point, this term has to be defined by characterising it accordingto the following scenario: E-learning deals with running an online course entirely virtually over a certainperiod and aims at mediating a set of competencies by means of objectives, learning materials and instruc-tions. All interactions between the learners (students) and the instructor (teacher) are accomplished onlineutilising an e-learning platform. Further, the assessment of the knowledge transfer as well as grading isalso conducted online.

Implementing e-learning courses can be seen as a complex process going beyond systematically ex-ecuted steps within an instructional design model. Among a large number of critical aspects, [McLeod,

27

28 3. Technology-Based Learning and Teaching

2003] suggests instructors consider principles of learning by means of historically grown learning theo-ries. Thus, it is possible to reuse certain procedures, for instance pre-defined instructional components asstated in [Merrill, 2001]. Within the e-learning situation, three learning theories – the Behaviourism, theCognitivism and the Constructivism – are of importance as shown in [Cooper, 1993], [Dietinger, 2003,p. 30ff], etc. In the following, these three theories are described in short and implications for realisingonline courses are derived.

3.1.1 Behaviourism

The behaviourist school of thought influenced by researchers like Pavlov, Thorndike, Watson and Skinner,who outlines that “learning is a change in observable behaviour caused by external stimuli in environ-ment” [Skinner, 1974, p. 2]. Behaviourists see “the mind as a black box, in the sense that a responseto a stimulus can be observed quantitatively, totally ignoring the effect of thought processes occurringin mind”. [Atkins, 1993] highlights four aspects relevant to realising online courses with respect to thebehaviourist school:

• The learning material should be divided into smaller instructional steps being delivered in an in-tuitive way by starting with a theoretical entity (a definition, category, rule, formula or principle),giving positive examples to reinforce mastery if the subject and showing negative examples to es-tablish conceptual boundaries.

• Course designers have to define sequences of instructions using conditional or unconditionalbranching to other instructional units and pre-determining choices within the course. In general,activities are sequenced by means of increasing difficulty or complexity. Further, a learner shouldcontrol the sequence and pacing through the materials.

• To maximise learning efficiency, learners should be routed to leave out or repeat instructions basedon their performance, which can be assessed by diagnostic tests within the sequence of learningactivities. Nevertheless, the instructional designer may also allow a learner to choose the nextinstruction out of a set of activities, giving the learner more control over the learning process and,thus, shifting learning to the paradigm of Cognitivism.

• The behaviouristic paradigm of learning suggests to demonstrate the required operation, procedureor skill and to break this instruction down into small parts, enriched with appropriate explanation.Thus, learners are expected to master the desired behaviour and acquire knowledge as well as skillsfrom frequent review or revision and by applying check tests or repeating practice with feedback.Instructional design focus on a low error rate and the usage of loops back through material if neces-sary. Furthermore, reinforcement messages should be used to maintain motivation.

Overall, behaviourists recommend a structured, intuitive approach to designing an online course, sothat basic concepts, skills and factual information can rapidly be acquired by the learners. Further im-plications regarding online learning can be summarised by the concept of drill and practice, portioningmaterials and assessing learner’s achievement levels and giving external feedback. However, the appli-cation approaches considering behavioural design is unproven or ineffective for higher-order learningactivities or for meta-cognition.

3.1.2 Cognitivism

Cognitivists consider learning as “an internal process that involves memory, thinking, reflection, abstrac-tion, motivation, and meta-cognition”, as outlined by [Ally, 2004]. Cognitive psychology comprises the

3.1. Relevant learning theories 29

learning process from an information processing point of view, where information is received in the sen-sory store through different senses and, further, transferred to the short-term and the long-term memorythrough different cognitive processes.

Furthermore, the cognitive school focus on individual differences between learners and suggests diver-sification of learning strategies to accommodate these differences. Thus, different learning styles – [Kolb,1984], [Myers, 1975], etc. – refer to how a learner perceives, interacts with and responds to learningmaterial. In addition, cognitive styles as addressed for example in [Witkin et al., 1977] describe preferredways of learners’ processing of information, i.e. modes of remembering, thinking, or problem solving.

Besides, the individual cognitive trend derived from Piaget’s theory, [Deubel, 2003] states that thelearning process also includes, according to Vygotsky, “socio-cultural perspectives emphasising sociallyand culturally situated contexts of cognition” [Duffy and Cunningham, 1996]. Such findings influencedinstructional design, for example by means of commonly-known instructional design theories like the“Cognitive Load Theory” [Sweller et al., 1998] or the “Cognitive Theory of Multimedia Learning” [Mayer,2001, p. 41ff]. According to these cognitive approaches, instructional designers have to consider thefollowing aspects for realising online courses:

• Teaching should enhance the learning process by addressing all sensors, focussing on the learner’sattention by highlighting important information, reasoning each instruction and meet the cognitivelevel of the learner.

• The instructional designer should tie up new information with existing information from long-term memory using advanced organisers to activate existing cognitive structures or to considerdetails of a lesson, providing conceptual models to enable the learner to retrieve existing mentalmodels, using pre-instructional questions to set expectations and to address the learner’s prior expe-riences.

• The learning content should be broken down into chunks to prevent cognitive overload. Exceedinga number of five to nine items to learn, linear, hierarchical, or spider-shaped information mapsshould be provided.

• Strategies requiring the learner to apply, analyse, synthesise and evaluate should be used to promotedeep processing of information and higher-level learning.

• It is recommended that learning materials include adequate activities and the right type of instruc-tional support for students with different cognitive and learning styles.

• With respect to dual-coding theory depicted in [Paivio, 1990], information should be presented ina multimodal way to compensate individual differences in information processing and to enhancethe transfer to long-term memory.

• Students need to be motivated to learn by means of learning strategies addressing the intrinsic(learner-driven) motivation and the extrinsic (instructor-driven) motivation. Therefore, methodssuch as Keller’s ARCS model – the abbreviation for attention, relevance, confidence and satisfac-tion [Keller and Suzuki, 1988] – could be applied by the instructor.

• With respect to [Mayer, 1998], the teaching strategy should enforce learners to use their meta-cognitive skills by reflecting on what they learn, collaborating with other learners or checking theirprogress.

• Finally, the teaching strategy should also connect learning content with different real-life situations,so that the learners can tie up to own experiences and, therefore, memorise things better. Further-more, the development of personal meaning and contextualisation of learning materials is supportedby a transfer to real-life situations of the information.


To sum up this subsection, cognitive psychology focuses on learners’ receiving and processing ofinformation to transfer it into long-term memory for storage. Therefore, instructional designers have toconsider different aspects beginning with chunking the learning content into smaller parts and supportingdifferent learning styles up to higher concepts such as motivation, collaboration or meta-cognition. Al-though the cognitive-focused approach is well suited for reaching higher-level objectives, a major weak-ness can be identified, if a learner lacks relevant prerequisite knowledge. To account for this, a coursedesigner has to ensure that the instructions are appropriate for all skill levels and experiences, which isevidently costly and time-consuming.

3.1.3 Constructivism

The constructivist school of learning suggests that “learners construct personal knowledge from the learn-ing experience itself” [McLeod, 2003]. Thus, learning can be seen as an active process and knowledgecannot be received from someone else or from outside. According to [Duffy and Cunningham, 1996],learners should be motivated to construct knowledge rather than being taught through instructions. Fur-thermore, constructivists emphasise situated learning, which emphasises learning within a certain contextand suggests strategies promoting multi-contextual learning to make sure that learners can apply knowl-edge broadly.

With respect to [Dimock and Boethel, 1999, p. 5f], the following assumptions can be made regard-ing this learning theory: Learning is an adaptive activity and situated in the context where it occurs.Knowledge is constructed by the learner who also deals with resistance to change. Experiences and so-cial interactions play a role in the learning process. By deriving implications for creating instructions foronline learning, the following statements have to be made up:

• Learning should be an active process. Therefore, students should carry out high-level activities,such as asking learners to apply information in practical situations, discussing topics within a group,asking for personal interpretation of learning content and so forth.

• To enforce learners constructing their own knowledge, instructors have to provide professionallycreated, interactive instructions, since students have to show initiative to learn, to interact withothers and to control the learning agenda [Murphy and Cifuentes, 2001]. Contrary to traditionallecture where instructors contextualise and personalise information to meet their needs, studentshave to experience the learning content at first-hand.

• As stated for example in [Hooper and Hannafin, 1991], collaboration and co-operation should beencouraged in the learning process. Students experience a real-life situation while working withothers, which facilitates the usage and improvement of their meta-cognitive skills in the following.Grouping learners for a collaborative work should be according to expertise levels and learningstyles, so that team members can benefit from one another’s strengths, as also approved with theimportance of the expert role in online discussions [Hasebrook and Maurer, 2004, p. 103].

• Learners should have control of the learning process. Besides, some kind of guided discoveryshould be provided in order to allow learners to make their own decision on learning goals, but alsoto require the instructor’s guidance and feedback.

• In the learning process, students should have enough given time and opportunity to reflect onthe learning content. Questions on the content embedded in the course can be used to encouragereflection and processing of the information.

• Learning should be made meaningful and illustrative by including examples and use cases fortheoretical information. Besides, activities should enforce learners to apply and personalise thelearning content offered.

3.2. E-pedagogy 31

• Instructors should focus on interactive learning activities to promote social presence and higher-level learning and to help develop personal meaning. As learning focuses on developing new knowl-edge, skills and attitudes, e-learning faces the problem that psychomotor, affective and higher-levelobjectives are hard to reach within online learning phases. Therefore, [Modritscher and Sindler,2005] suggest providing other ways – such as social or interactive activities, context-based learning,assessment through open-ended questions, etc. – to realise such didactical aspects.

Examples of constructivist learning can be found within the scope of experiential learning, self-directed learning, context-aware learning and reflective practice. Despite a variety of advantages to Con-structivism, like the presentation of content from multiple perspectives, the active knowledge construction,the development of meta-cognitive strategies, this learning theory also faces a few disadvantages, such asproblems in adequately evaluating the learning process, lack of instructional resources to respond to themultitude of student interests or higher effort to create context-based learning content, restrictions on driv-ing the learning process to a certain direction given for example by science, higher drop-out rate due to alack of extrinsic motivation for students with low capabilities for self-directed learning, etc.

These three commonly-known learning theories are of central relevance for examining pedagogicalissues and the implementation of different e-learning strategies, as shown in the following two sections.

3.2 E-pedagogy

Referring to [Knowles et al., 1998, p. 10], education can be understood as “activity undertaken or initiatedby one or more agents that is designed to effect changes in the knowledge, skill, and attitudes of individ-uals, groups, or communities”. On the contrary, the term “learning” emphasises the person in whom thechange occurs or is expected to occur. Thus, learning comprises “the act or process by which behaviouralchange, knowledge, skills, and attitudes are acquired” [Boyd and Apps, 1980, p. 100f]. Tying up to thisdefinition, the following subsections deal with the psychological and learner-centred aspects of the tra-ditional learning process – such as relevant factors for learning, characteristics of learners and furtherinfluences on learning – and examine them within the context of e-learning.

3.2.1 Factors relevant to the learning process

Drawing conclusions from [Bransford et al., 2000, p. 51ff], four factors can be outlined as significantlyimportant for the learning process: (1) attention, (2) motivation, (3) emotions and (4) experiences of thelearner.

First of all, the focus of attention determines if a student mentally follows a lecture and, therefore,if the intended behavioural change affects a learner at all. E-learning particularly requires a strategy forgetting and keeping the learner’s attention. Thus, it is necessary to consider cognitive processes such as thelearner’s selection of incoming data into the sensory memory, organising and integrating this informationby building connections in short-term memory and encoding it by transferring it to long-term memory.Thus, it is recommended to apply certain principles for instructional design, for example the ones by[Fleming and Levie, 1993].

Secondly, the motivational states of students are of importance when questioning how the stimuligiven by the teacher promotes the learning process. [Bransford et al., 2000, p. 60] state that “motivationaffects the amount of time that people are willing to devote to learning”. Yet, this willingness to learn iscaused by different motives, beginning with the intention of achieving something while competing againstcolleagues, or helping other people, up to emotional factors like anxiety. [Entwistle, 1981] classifiedthree motivational orientation styles: (a) meaning-oriented, (b) reproducing-oriented and (b) achieving-oriented motives. Considering motivational aspects for e-learning is mainly dependent on the learning


content itself, for example by pointing out the relevance for an instruction or including multimedia andinteractive elements such as games and simulations, as shown by the TRIANGLE software [Holzingeret al., 2006]. Furthermore, it is advantageous to create competition within a learner group and adaptto pre-knowledge in the subject domain to prevent the students from being unchallenged. For instance,[Astleitner and Keller, 1995] describe a framework for adapting instruction to the learner’s motivationalstate in computer-assisted instructional environments.

Thirdly, emotions have, similarly to motivation, a strong impact on the learning process, as outlinedby [Hasebrook and Maurer, 2004, p. 32]. [Tobias, 1987] points out findings on students’ performancedepending on anxiety, in particular test anxiety and proposes special methods for dealing with such prob-lems. On the other side, an emotion – no matter whether a negative and positive one – may influencelearning due to its special nature. With respect to [Paulsen, 2005], “emotion is an unconscious arousalsystem that alerts us to potential danger and opportunities”. Thus, addressing a learner’s emotional chan-nel can be seen as a key cognitive process for transferring data into the short or even long-term memory.Within e-learning the improvement of the learning process can be realised through emotions, for exampleby storytelling, provocation, emotional figures and animations, group work, enabling confidence in thelearning content, etc.

Fourthly, knowledge transfer can be improved if learners can tie up to prior knowledge either inthe same domain or in a similar context. [Slavin, 2006, p. 181] states that “interference happens, wheninformation gets mixed up with, or pushed aside by, other information”. At the beginning, the degree ofmastery of the original subject influences the learning process [Bransford et al., 2000, p. 53]. In particular,an adequate level of initial learning is required. Learners can then construct new understanding by tyingup to previous experience which may not have been activated yet. In this way, learners become capableof understanding conceptual changes, adopt knowledge regarding their culture or everyday life and evenimprove meta-cognitive abilities.

Research findings have shown that the higher the level of prior achievement within a domain or acontext, the less instructional support is required to accomplish a task [Tobias and Ingber, 1976]. Referringto [Tobias, 1994], prior knowledge strongly relates to interest in the subject. Considering prior knowledgewithin online courses, the macro-adaptive instructional approach described in [Park and Lee, 2004] dealswith the necessity to determine learning objectives, to define dependencies between instructional units, andto assess the students’ competencies to grant access to restricted instructions. These aspects are highlydependent on the learning content so that well-established e-learning standards – such as the specificationsof SCORM [ADL, 2004] – fulfil these requirements.

3.2.2 Learner characteristics

Drawing conclusions from the last subsection, a strong impact on learning is given by the individualdifferences among learners, as stated for example by [Cronbach, 1957]. According to literature, eachlearner differs from another by means of the following aspects, so-called learner characteristics:

• First of all, each learner has a unique profile of intellectual capabilities, which can be characterisedby Gardner’s Multiple Intelligences [Gardner, 1993] or various types of cognitive abilities describedin [Corno and Snow, 1986]. Education deals with the theory of multiple intelligences in two ways:On the one side, teachers devise curricula addressing different intellectual capabilities. On the otherside, educators focus on the development of specific intelligences, of intra or interpersonal skills forexample. Although it is rather unmanageable to consider the learners’ intellectual abilities withinthe classroom or e-learning situation, [Kelly and Tangney, 2003] applied Gardner’s theory withinan intelligent tutoring system named EDUCE.

• Secondly, learning preferences usually result from predispositions or orientations to learning and

3.2. E-pedagogy 33

can be seen as influences by the context [Jarvis and Woodrow, 2001]. [Dunn et al., 1989] classifypreferences by four different areas: (a) environmental, (b) emotional, (c) sociological and (d) phys-ical. Preferences are considered by many e-learning environments in various ways, for exampleby adapting the language or presentation of the learning content, group models, etc. Exemplarysystems can be found i.e. in the field of adaptive educational systems [Brusilovsky, 1996].

• Thirdly, researchers in the field of learning and teaching introduced so-called cognitive and learn-ing styles which are somehow related to intellectual capabilities and preferences. Both kinds ofstyles try to provide more practical models for teachers. Cognitive styles, such as field-dependence,reflectivity versus impulsivity, haptic versus visual and so forth, characterise modes of perceiving,remembering, thinking and decision making. Learning styles like holist versus serialist, percep-tion styles, concept formation approaches, etc. try to describe the connection between instructionalpresentation and materials with a student’s preferences and needs [Schmeck, 1988]. Overall, manypractical models like the WAVI model by [Riding, 1991] – for example applied within the AdeLEproject [AdeLE, 2006] – or the learning styles by [Kolb, 1984] – realised in the AHA! System [Stashet al., 2004] for example – have been developed in the last decades.

• Fourthly, [Modritscher et al., 2004c] highlight constitutional attributes and states of learners,which may deal with physical properties of the body like disability, age, amblyopia, etc. as wellas with short-term states of students, such as tiredness, concentration, emotional and motivationalstates and the like. Both directions are already well-examined and various systems try to consideraspects of physical properties – for instance disabilities as stated in [Sanchez and Flores, 2004] – orconstitutional states of learners such as the learner’s attention [Ueno, 2004].

• Fifthly, self-efficacy and meta-cognition influence the learner’s achievement in the learning pro-cess [Bandura, 1982]. Self-efficacy comprises a student’s evaluation of the ability to perform agiven task through different senses. Furthermore, meta-cognition stands for the awareness of theprocess of learning and consists of two basic processes (see [Nelson and Narens, 1994], [Winn andSnyder, 1996] or [Hasebrook and Maurer, 2004, p. 96]): (a) monitoring the learning process and(b) adapting the learning strategy. According to [Park and Lee, 2004], meta-cognitive abilities ex-amined by various researchers within the area of aptitude-treatment interaction (ATI) are closelyrelated to the learners’ experiences and have an impact on different variables, such as the degree offeedback and tutoring, the locus of control, personality attributes and so forth. In particular, varioussystems in the scope of adaptive hypermedia – for example by methods like adaptive navigationsupport [Brusilovsky, 1996] – focus on learner control.

• Sixthly, the background knowledge of a learner comprising language and computer skills as wellas experience on a related situation by means of a familiar context may also have an impact onlearning. For example, [Campbell et al., 2004] report that students from abroad may have problemswith understanding the language. [Felder and Henriques, 1995] examined learning styles within thescope of language education and found out connections with learning styles. Thus, [Modritscheret al., 2005] suggest providing translations for problematic phrases to support the learning process.Anyway, various approaches in the field of e-learning focus on experience of work in related areas,the user’s profession, experience of using the platform (e.g. see [Brusilovsky, 1996]) as well as onforeign language students (e.g. see [EPHRAS, 2007]). Further, [Akhras and Self, 2000] introducethe INCENSE system offering the ability of identifying and analysing different learning situationsand, if necessary, automatically switching among them.

• Finally, the last and most relevant characteristic of learners involves the user’s prior knowledgeand experience in the domain. [Vassileva, 1996] differentiates between experience and real knowl-edge about a topic, where experience determines the user’s model of a knowledge space, the way


of browsing through and mastering tasks in a domain, etc. Similarly, [Bransford et al., 2000, p. 31]state that people having developed expertise in a particular area are able to think effectively aboutproblems in this area and, therefore, differ from novices. Both factors are relevant to the learningprocess as shown in the last section. Thus, most theoretical models – for example the macro-adaptive instructional approach dealing with adaptation of the learning process according to the stu-dent’s pre-knowledge and dependencies between instructional units [Park and Lee, 2004] – and sys-tems – like AHA!, ELM-ART, Interbook, KBS Hyperbook, PT and so forth [Henze, 2000, p. 16ff]– focus on the prior knowledge of a learner. Experience is considered within the field of variousresearch fields such as adaptive hypermedia, for example by methods like adaptive navigationalsupport for browsing through a hyperspace [Brusilovsky, 1996].

Concluding these seven classes of learner characteristics, it can be said that the result of learning ishighly dependent on the learners themselves. Therefore, teachers as well as e-learning content creators andonline instructors have to know very much about pedagogical aspects and provide a large set of methodsto support different kind of learners by means of the characteristics depicted above. Nevertheless, thereare still more influences on learning as shown in the upcoming subsection.

3.2.3 Further influences on learning

While the last subsection dealt with learner-dependent attributes affecting the learning process, somefurther aspects – given by didactics or related to learner characteristics – may have an impact on learning.The following paragraphs therefore examine various factors influencing learning.

Referring to [Krapp et al., 1992], interest in a certain domain depends on aspects of prior knowl-edge, emotions and motivation, all of them treated in the last two subsections. Thus, it can be assumedthat interest results from expertise enhancing the degree of the learner’s self-confidence as well as for-mer positive experiences with the subject and can be modelled on the bases of learner characteristics bymeans of inferring a certain amount of interest derived from these factors. As far as supported by thedidactical strategy, e-learning may consider the factor interest by providing adaptability, for example byallowing the learner to choose preferred learning content, implementing statistical methods to determineinstructions which are more interesting for the learner, or applying taxonomies and IR-based strategiesto adapt the learning process on the basis of the factor interest. Examples of platforms comprise a widerange of systems beginning with the ones focusing on motivation or prior knowledge, others dealing withself-organising hypertext maps like the KnowledgeSea [Brusilovsky and Rizzo, 2002] or environmentssupporting constructivistic learning, for example the idea of navigating on the basis of a course’s conceptsas mentioned in [Modritscher et al., 2005].

A didactical key issue is about people remembering and forgetting. While these processes are closelyrelated to intellectual capabilities, meta-cognition, prior knowledge and motivation of learners, a teachermight antagonise the forgetting curve, for example the one introduced in [Ebbinghaus, 1963], by regu-larly repeating relevant content. To define such key concepts within a course, teachers often use adequatelearning objectives and choose appropriate teaching strategies. Such issues are considered within variouse-learning environments or even specifications for learning content, for example by typical concepts of themacro-adaptive instructional approach examined in [Park and Lee, 2004], by competency-driven strate-gies like the knowledge space theory [Falmagne et al., 2003] or even by the possibilities of instructionalsequencing within the specifications of SCORM [ADL, 2004].

Another didactical aspect deals with the time to learn or the so-called time on-task. [Bransford et al.,2000, p. 42] point out the necessity of giving the student enough time to reach an appropriate level ofexperience within the scope of a certain domain, in particular for a complex subject matter. Nevertheless,learners are often faced with the situation in which a teacher tries to cover too many topics too quickly,which hinders an effective knowledge transfer for different reasons. On the other side, the time on-task

3.3. E-didactics 35

should be limited, so that the learner is sufficiently challenged and self-efficiency is able to increase.Thus, it is particularly important for e-learning to plan the time allocated for learning and the time reallyspent on learning. As pointed out in [Dietinger, 2003, p. 31], it is one of the advantages of e-learningthat learners can go through the course materials at their own pace. Thus, deadlines must be realistic inorder to avoid frustrating of the students. The possibility to define and manage deadlines for instructionalunits is provided by the commonly-known specifications for e-learning content and nearly each learningmanagement systems, even by open source solutions such as Moodle [Moodle.org, 2007].

Depending on the given learning objectives, issues like feedback and tutoring might be of relevancefor the learning process. In particular, if a course aims at mediating high-level objectives, skills or a certainbehaviour, it is important for successful learning to give immediate feedback (e.g. see [Thorndike, 1913]).With respect to [Park and Lee, 2004], various research areas, such as aptitude-treatment interaction or themicro-adaptive instructional approach, deals with giving feedback and technical solutions like intelligenttutoring systems or techniques for natural language dialogues. Furthermore, [Modritscher and Sindler,2005] suggest the application of methods such as simulations, games, automatically essay grading, quizzescreated with professional authoring software and so forth.

Finally, learning is also affected by the context in which knowledge transfer takes place. Accordingto [Bransford et al., 2000, p. 43], learners might be able to learn in a certain context, but fail to learn inanother one or to transfer the experiences gained to other contexts. Contextualised knowledge is regardedonly by few e-learning environments – one of them is the INCENSE mentioned in the last subsection. Asthis issue is closely related to constructivistic theory, new paradigms are of importance nowadays. Oneidea in this scope is the application of a Dynamic Background Library [Dietinger et al., 1999] to supportcontext-driven learning [Modritscher et al., 2005].

Concluding this section, it has to be pointed out that the issues depicted so far comprise just the mostrelevant and learner-centred factors of the learning process. A full overview of the complexity of learningcan be read in [Bransford et al., 2000, p. 31ff], for example. Nevertheless, it can be stated that the mostcritical factor for successful learning is the learner. The most important difference between the classroomsituation and e-learning can be outlined with the statement that a teacher can adapt the learning processmuch more effectively by holding a lecture in the class, since communication in both directions – from theteacher to the learners and vice versa – is faster and more effective. Contrary to this, it is much harder toevaluate a factor of the learning process or learner characteristic and react to it via an e-learning platform.Therefore, research streams such as adaptive instructional systems or adaptive e-learning [Park and Lee,2004] deal with aspects of adaptation in e-learning environments to improve the learning process.

3.3 E-didactics

Beside the pedagogical issues within the learning process, the viewpoint of teachers also plays an impor-tant role for implementing online courses. Although cognitivists and particularly constructivists do notbelieve that learning can be driven from outside (see also section 3.1), educators are able to give stimuli toeffect changes in knowledge, skill or attitudes of learners, as outlined at the beginning of the last section.Moreover, didactics do have an impact on the learning process, as indicated with the learning theoriesrelevant for e-learning at the beginning of this chapter and shown by means of a case study in chapter 10.Therefore, this section deals with didactical aspects of e-learning.

Teaching itself – no matter if in the classroom or via a learning platform – is a very complex task.[Bransford et al., 2000, p. 191f] states that a teacher must not only cope with a course’s subject matter,but also master different didactical skills to plan and run a course successfully. Backed up by literature[Bransford et al., 2000, p. 131ff] and practical guidelines [IDS, 2002], the simplified didactical process inthis dissertation consists of the following four stages: (1) the didactical planning, (2) the implementationof a course, (3) the assessment of the learning process and (4) the course evaluation and revision. In the


following subsections, these four stages are examined for the online teaching process by comparing themwith teaching in the classroom situation.

3.3.1 Didactical planning

Planning an online course comprises different tasks, beginning with collecting and assembling learningmaterials, analysing the target group, defining learning objectives as well as determining learning activitiesand assessment methods. Learning objectives are important to specify which competencies should bemediated to the learners and to which extend should these competencies be mastered.

Learning objectives mainly depend on two issues: On the one side, it is necessary to consider param-eters given by the organisation, for example the title of the course or crossovers to other courses. On theother side and in accordance with [IDS, 2002, part II], the prior and the background knowledge of thelearners should be addressed by means of a didactical analysis on the basis of aspects of the last section,particularly the learner characteristics.

With respect to didactics, the first step of the teaching process deals with the kind of competenciesto be mediated to the students. Therefore, [Durand, 2000] made up a theory on the basis of HowardGardner’s Multiple Intelligences and describes three main classes of competencies:

• First of all, knowledge can be seen as a kind of mental model about parts of the real world. Inother words, knowledge corresponds to a number of facts stored in an individual’s memory andconnected to other pieces of assimilated information. This dimension can be denominated as thecognitive dominion.

• Second, a competency can also be a skill, which is related to the capacity of applying and usingacquired knowledge. According to [Bloom, 1956, p. 38f], a skill can be seen as process wherean individual uses appropriate techniques and information in order to examine or solve a problem.Skills can be divided into intellectual skills, which are about mental processes manipulating infor-mation and psychomotor skills, where a neuromuscular coordination is performed.

• Finally, attitudes are concerned with social or affective aspects. [Petry et al., 1987, p. 15] considerattitudes to be “complex mental states of human beings that affect their choice of personal actiontowards people, things and events”. An attitude can be seen as a feeling, emotion or a degree ofacceptance or rejection of a person to other persons, objects or situations.

A competency in practice is supposed to consist of more than one of these classes. In most casesa strong focus on one of these three classes can be recognised, but there can also be adequately mixedcompetencies.

After considering what should be taught within a course, it is important to decide to which extent andunder which circumstances the competencies should be mastered by the students [Hasebrook and Maurer,2004, p. 134]. Therefore, a teacher has to define learning objectives following some kind of taxonomy, forinstance the one by [Bloom, 1956]. For the cognitive domain, the different levels of objectives are thefollowing ones:

• The lowest level of objectives is about recognising and recalling assimilated information.

• Based on these abilities, a student can comprehend and explain what he internalised.

• In the next step, the gained knowledge can be applied in new situations.

• At the analysis level, the student is able to analyse, structure and organise the facts and concepts.

3.3. E-didactics 37

• Synthesis describes the ability to reassemble the pieces of assimilated information to create newknowledge.

• At the highest level, a student can even evaluate the value of ideas and cognitive materials.

Level Cognitive Domain Psychomotoric Domain Affective DomainI Knowledge Imitation ReceivingII Comprehension Manipulation RespondingIII Application Precision ValuingIV Analysis/Synthesis Articulation OrganisationV Evaluation Automation Practicing what you preach

Table 3.1: Bloom taxonomy [Bloom, 1956], adapted and extended for skills and attitudes

These different levels of learning objectives can be also made up for skills and attitudes, as shown intable 3.1. Furthermore, it is possible to define additional conditions for each objective, such as the usageof a tool or the time extent. In general, the procedure for creating the learning objectives for a course startswith defining very abstract objectives which are broken down to the detail subsequently.

In practice, cognitive psychologists examine learning on the basis of competencies, relations betweenthem and learners’ knowledge states. For instance, [Albert and Hockemeyer, 1997] highlight hypertext-based learning with respect to the so-called “Knowledge Space Theory”, a formal approach to modelcompetencies and their interdependencies. [Albert and Hockemeyer, 2002] report on applying the knowl-edge space theory for creating and adapting the course structure, for adaptive assessment of the learners’knowledge state and for adaptive training.

The stage of didactical planning is necessary and equivalent for both traditional teaching and dis-tance education. Moreover, learning objectives have an impact on the instructional design as well as onassessment of the learning process as pointed out in the next two subsections.

3.3.2 Implementation of a course

After the didactical planning of the course, a teacher has to deal with the following tasks. On the oneside, organisational and administrative requirements, such as the arrangement of the infrastructure, thedetermination of the schedule, the provision of the learning materials, etc., have to be considered. On theother side, a teacher has to select and determine the course’s curriculum. Contrary to didactical planning,this stage is characterised by significant differences between traditional teaching and online courses.

As already indicated with the instructional strategies based on the commonly-known learning theoriesin section 3.1, instructional design is a very broad field of study and can be found in various other sources,for example in [Bransford et al., 2000, p. 131ff], [Jonassen, 2004], [IDS, 2002, part III] and so forth.Nevertheless, a few statements about instructional design have to be manifested here:

• An instruction generally consists of a task assignment (activity) and learning material (knowledgeartefact). Instructions can be created by a teacher or a professional content creator, for exampleapplying own tools, or retrieved from a certain repository, for instance from a learning contentrepository or the Internet. Often, teachers prefer creating their own instructions, although it is morecostly and time-consuming than reusing existing instructions or delegating this task to a professionalcontent creator, as stated for example by [Lennon and Maurer, 2003].


• Instructions are delivered according to a sequence given by the teacher or randomly accessed bythe learner. There might also be dependencies between instructions, i.e. a learner has to completesome instruction to access another one.

• The implementation of an online course – particularly the instructional design and sequencing– mainly depends on the learning objectives determined within the didactical planning stage.While [Bransford et al., 2000, p. 134] differentiates between learner, knowledge, assessment andcommunity-centred learning design, younger research streams address the relation between com-petency types (learning objectives), learning activities and instructional strategies, which led to thefollowing approaches: (1) guidelines for choosing appropriate teaching methods according to givenlearning objectives [Bransford et al., 2000, p. 136], (2) standardisation efforts for learning designand didactical patterns [IMS, 2007d] and (3) workflow-based learning processes [Helic, 1995].

• From a technological viewpoint, this stage – as well as the assessment of the learning processexamined in the next subsection – is implemented within a learning platform. Therefore, it is rec-ommended that such a system, a so-called Learning Management System (LMS) [Dietinger, 2003,p. 41ff], provides many features in order to support a wide range of didactical strategies.

Holding lectures virtually offers a few advantages. Amongst others, [Dietinger, 2003, p. 23ff] outlinesthat the learning process can be enhanced using visual and interactive content. Moreover, an online in-structor might not require all the soft skills which are necessary for a teacher in the classroom situation.However, an e-teacher has to deal with other competencies, like the usage of the learning platform, mediaskills, etc. Further, [Modritscher et al., 2006b] showed that it is difficult to mediate affective, psychomotorand any kind of higher learning objectives via distance learning. Finally, the assessment of the learningprocess is harder within the e-learning situation as shown in the following.

3.3.3 Assessment in the e-learning situation

[IDS, 2002, part IV] points out the necessity of assessment which should be executed not only to gradestudents, but also to measure the learning process. In addition, assessment methods have a great impacton the students’ learning behaviour as stated by [Scouller, 1998]. Generally, the assessment of learningis divided into two processes: On the one side, teachers should apply methods of formative assessmentin order to enhance the learning process and evaluate the factors of learning mentioned in the last section[Bransford et al., 2000, p. 19]. On the other side, the achievement of the learning objectives defined bythe teacher has to be assessed by means of determining a mark (summative assessment).

Within traditional educational styles, a teacher might apply various methods, beginning with oralquestionnaires up to written exams. For online courses, assessment is often reduced to limited-choice oropen-ended questions. Limited-choice questions such as multiple choices are applied to reach lower-levelobjectives like recalling facts. Open-ended questions like sentence completion, short answers, essays etc.require students to formulate their own answers which do not have to be pre-determined. Referring to[Scouller, 1998], open-ended questions can be used to evaluate higher-level objectives like applying orevaluating assimilated knowledge.

Teachers have to consider which type of question they use for assessment depending on the level oflearning objectives, size of the class, reliability in grading, prevention of cheating, exam construction andgrading time and several other criteria. When examining the didactical aspects treated so far, the followingproblematic areas can be identified within the e-learning situation:

• All kinds of competencies – knowledge, skills and attitudes – may be mediated within an e-learningenvironment. Therefore, it is possible to create learning content including facts relevant to a learner,instructions how to achieve a skill, or information about an expected behaviour. Thus, technology

3.3. E-didactics 39

can be seen as an enabler for these types of competencies, because information can be enriched withmultimedia assets [Gunawardena and McIsaac, 2004], practicing skills can be supported by usinginteractive elements or tutoring systems and the behaviour of a student can be observed within thecontext of the e-learning system in terms of the micro-adaptive approach for e-learning [Park andLee, 2004]. In fact, it is easier to mediate knowledge through e-learning environments, while theeffort for teaching skills or attitudes is much higher as shown in the case study in the next section.

• Within an e-learning system, objectives need to be defined regarding the target group. With respectto the standardisation process in the field of e-learning, specifications such as SCORM alreadyallow description of objectives as meta-information for the course [ADL, 2004]. Nevertheless, anobjective specified with SCORM can be seen as a state within the system and does not tell anythingabout the level of the learning objective. Furthermore, it is hardly possible to reach high-levellearning objectives for all three types of competencies within a pure e-learning situation as statedlater in the study.

• Learning objectives which are defined by a teacher always have to be evaluated in some way – tograde the students and to improve the quality of the course for future sessions. Considering thepossibilities of e-learning, it is well documented that knowledge acquisition can be assessed by us-ing limited-choice questions like quizzes or multiple-choice questions. Nevertheless, for most areasand, in particular, to reach high-level learning objectives it is necessary to examine students by ask-ing open-ended questions, as reasoned for example by [Scouller, 1998]. Furthermore, the answersto such questions have to be interpreted and evaluated by experts. Researchers try to imitate suchexpertises using artificial intelligence methods within intelligent tutoring systems, but the results arestill rather limited [Park and Lee, 2004]. In terms of skills, the learning results cannot be measuredby technology-based methods without hard efforts.

It has to be outlined that the assessment of high-level objectives can be realised in many different ways.With respect to the assessment methods focusing on didactical aspects such as defining competencies andevaluating the learning process according to the determined learning objectives, the following possibilitiesfor implementing assessment in the e-learning situation can be found in the literature:

• First of all, most e-learning systems offer the possibilities to create and provide limited-choicequestions. Although quizzes can save a lot of time to grade a large amount of students and [Scouller,1998] reports on good results for low-level objectives of the cognitive domain, they show a worseperformance for the employment of deeper learning strategies and higher levels of cognitive pro-cessing.

• Therefore, [Scouller, 1998] states that it is necessary to implement open-ended questions withinthe e-learning situation, for instance by tasks like writing essays or submitting some sort of projectwork. It is obvious that the evaluation of such tasks is extremely time-consuming for a teacher.Therefore, it is recommended to apply supporting methods such as automated grading, for exampleusing the Markit c© system introduced in [Williams and Dreher, 2004].

• As an extension of automated essay grading, intelligent tutoring systems may provide some kindof expertise within a domain and allow fully automated teaching and assessment, as stated in [Parkand Lee, 2004]. Yet, this kind of system is hard to realise, often restricted to a certain domain and,thus, to a few learning objectives. An example of a rather complex system in this area is INCENSEproviding different scenarios for teaching of the software engineering process [Akhras and Self,2000].


• [Lennon and Maurer, 2003] describe several approaches beginning with the usage of professionalauthoring software up to a shift to the constructivistic learning paradigm. On the one side, auto-matically generated crossword puzzles may be an enabler for the students’ interest and motivationand have positive effects on assessing low to medium-level objectives of the cognitive domain. Onthe other side, applying constructivistic learning methods requires a high level of students’ self-motivation, but can reach high-level objectives in all domains as shown in the next few paragraphs.

• One aspect of constructivism deals with collaborative learning. In particular, group activitiesrequiring students to discuss a topic are a powerful element to extend the possibilities of e-learningas outlined by [Piaget, 1977, p. 157]. As a conclusion, students may treat open-ended questions,when they are working in groups.

• Another interesting concept of constructivism is a so-called peer assessment, which was applied asone assessment method in the case study comprised in the next section. As described in [Bhaleraoand Ward, 2001], peer assessment may reach high-level objectives for all possible domains andprovide other advantages, such as using natural language processing, lowering the effort for theteacher, etc.

• Finally, [Gredler, 2004] reports on applying games and simulations for e-learning which can be asuccessful approach to reach high-level objectives, in particular for intellectual skills, but also formediating knowledge or internalising value systems.

Nevertheless, assessment in the classroom is much more efficient, as the teacher can easily realisesome method and adapt the learning process according to the learners’ feedback. Thus, researchers inthe field of technology-based learning and teaching try to implement such behaviour within e-learningsystems, as shown in the next chapter.

3.3.4 Evaluation and revision of courses

After completing a course and marking the learners’ achievements, a teacher has to evaluate and reviseit in order improve their own teaching skills as well as the upcoming courses. With respect to [IDS,2002, part V], this stage of the teaching process can be realised by conducting formative and summativeevaluations, for example by using questionnaires, surveys, minute papers or other methods to get thelearners’ feedback. Evaluation and revision of courses are equal for both traditional teaching and e-learning and can be seen as an important issue within any kind of educational organisation for qualityassurance reasons.

3.4 Towards formalising e-learning and e-teaching

Concluding this chapter, this section sums up relevant research issues relevant to technology-based learn-ing and teaching. Therefore, an overview of these issues is given and a formal framework for learning con-tent, pedagogical issues and didactics is built up as a basis for examining the field of adaptive e-learningin the next chapter.

3.4.1 Overview of technology-based learning and teaching

As a consequence of the first three sections, the primary focus of technology-based learning and teachingis, as with traditional teaching, the learning process which is often also named “knowledge transfer”.

3.4. Towards formalising e-learning and e-teaching 41

Figure 3.1 visualises that learning depends not only on certain factors, such as the motivation, emotionalstates, the learner’s attention, etc., but also on certain characteristics of a learner and the didactical strategy.Examining the learning process, for example towards its efficiency, is comparable to systems theory in theway that the teacher can measure the learning outcome, i.e. by applying assessment or evaluation methods,in relation to the input, by means of the efforts for planning and implementing a course.

Figure 3.1: Overview of research issues related to the learning process

Besides evaluating the learning process from a didactical viewpoint, it is also interesting how cer-tain factors of learning or characteristics of the learner are related to or can be influenced by didacticalstrategies. Such research questions are examined not only by psychologists or pedagogues, but also intechnological driven streams like adaptive e-learning, as shown in the next chapter. Moreover, chapter 10reports on a study about the pedagogical and didactical impact of different e-teaching strategies.

However, at this point three relevant models for technology-based learning and teaching are intro-duced. The following section therefore deals with learning materials by means of the course’s content,while the further sections try to formalise pedagogical and didactical aspects of e-learning.

3.4.2 Content model

The learning content given as a set of artefacts, for example by a learning content repository, can be seenas the basis of the content model. These atomic entities of the content – whether available in digital formor as a reference to a real object – can be connected to certain domains and appear within determinedcontexts. With respect to [Gutl and Garcia-Barrios, 2005a] a domain can be modelled using concepts,while a context is defined as a set of situations. To keep the content model simple, a domain consists ofa set of concepts, a context of a set of situations and learning content of a set of artefacts (see figure 3.2).This model does not care about structuring a domain, a context or the content. Furthermore, artefactscannot be divided into sub-entities.


Figure 3.2: Formal specification of the content model

Hence, there are three functions relevant to this simple model: First of all, the relation between aconcept, a situation and an artefact has to be given by means of a certain degree of relevance. This relationfactor could be assigned by an expert, for example the teacher, or through IR-techniques. Secondly andthirdly, it is necessary to define dependencies between a pair consisting of a concept and a situation andcheck such dependencies recursively. Yet, it is not allowed to have a cyclic loop within these dependencyrelations.

In the case of e-learning the artefacts are supposed to be digital, the context could describe the level ofmastery for a concept – for example by using a taxonomy like one by [Bloom, 1956] – and the relevancebetween a concept and the artefacts as well as the dependencies between concepts have to be determinedby the teacher. In this context, allowing cyclic dependencies would cause the problem that the teachercould not chose an instructional entry point for a course, because each instruction would require at leastanother one and, therefore, dependencies could not be resolved.

An example of a content model would be an approach like the knowledge space theory [Albert andHockemeyer, 1997], which deals with the skill-model beyond a set of knowledge artefacts, whereby a skillcan be seen as a pair of a concept and a situation (e.g. an action verb of Bloom taxonomy of educationalobjectives). As the formal content model allows definition of relations between skills and artefacts and,further, dependencies between skills, it allows description of the knowledge space theory. Simpler exam-


ples of a content model would be a conceptual space of the given knowledge artefacts or also dependenciesbetween artefacts by means of a recommended chronological sequence to go through them.

3.4.3 Pedagogical model

The pedagogical model deals with factors influencing the learning process as mentioned in section 3.2.The specification shown in figure 3.3 allows defining user and environmental states that are fully dependenton a learner, a concept and a situation as well as any combinations of these three factors. Thus, it ispossible to determine states completely independent of the learner (domain or context-related states) orfully dependent of the learner (characteristics and user states).

Figure 3.3: Formal specification of the pedagogical model

The mapping in the pedagogical model can be generated and modified in the following way: Onthe one side, environmental and learner-relevant factors have to be determined a priori with respect tothe needs of an e-learning system. On the other side, the mappings might be adapted later on. Thepedagogical model can be used to characterise or adapt the learning process, for example by applyingappropriate sensors, measuring certain states (using the operation “AssessState”) and reacting somehowto modified states. Considering all relevant factors of the learning process, pedagogy can be seen as theunion of environmental and learner-specific states.

To give an example of a pedagogical aspect, environmental properties might be defined by a mappingfrom the triple “any student”, “any concept” and “a specific property of the surrounding” (e.g. the levelof the background noise) to a certain level (e.g. given in dB). Prior knowledge of a student in a certaindomain can be seen as mapping of the triple consisting of “a student”, “a certain concept” and “the specificsituation” (e.g. “Student X can explain concept Y”) to the state “mastered” or “not mastered”. Intellectualabilities of a learner might be given as mapping of the student to the results of an IQ test.

Although pedagogy should consider all these aspects, the assessment of pedagogical states as well asthe learning platform’s possibilities are part of the next model, the didactical model.


3.4.4 Didactical model

The didactical model introduced in this subsection tries to consider the most relevant aspects of e-teachingoutlined in section 3.3. Therefore, the model deals with determining learning objectives, instructionaldesign as well as assessment methods for e-learning courses. Nevertheless, some simplifications pointedout in the next few paragraphs had to be made due to the high degree of complexity of teaching, whichrequires not only fundamental knowledge of the course’s domain, but also a large set of practical skills.

Figure 3.4: Formal specification of the didactical model (types and instance variables)

In detail, e-didactics consists of learning content, pedagogical aspects (both given by the models fromthe last two subsections), a set of learners, the features provided by the learning platform, a sequence ofobjectives, a sequence of instructions (the course itself) as well as information about mastering learningobjectives, the learning progress and the relevance and suitability of instructions (see figure 3.4). Havingthese sets of basic instance variables allows description of an online course from a didactical viewpoint.

An exemplary scenario would be a course on a certain topic and realised for a group of learners usingan e-learning system. The online course would consist of a sequence of instructions and a set of objectivesto achieve, by means of a certain mastery level, whereby each instruction is relevant to one or moreobjectives. Within the online learning process the progress of the learners is measured according to theirmastery of the learning objectives. Pedagogical issues and their suitability for instructions are ignored forthis simplified example.


The operations of this model summarise the most relevant activities of a teacher planning and imple-menting courses (see figure 3.5). First of all, the teacher has to determine the learning objectives regardingdependencies given by the content model. Further, the minimum level to master the objectives has to beprescribed. Secondly, the instructor has to add instructions, which can be seen as a pair of an educationalactivity (e.g. a discussion, a task, a quiz, a poll, a certain navigational element, etc.) and an artefact, to thecourse. Furthermore, the teacher is free to specify other instructions relevant to a certain objective as wellas another instruction suitable for another pedagogical state. These aspects are of importance for adaptingthe learning process as explained in the next subsection.

Figure 3.5: Formal specification of the didactical model (operations)

Thirdly, the instructor has to assess the learning process in some way. As depicted above, this is somekind of a simplification of this model, because assessment may also be realised within an instruction, forexample with quizzes or automatic essay grading. Yet, such aspects would increase the complexity of themodel significantly. Finally, the last operation is about analysing the learning process (by the operation“IsMastered”). The question, if all learners have achieved all given objectives, can be seen as the central


target of didactics.

In accordance with [Spector and Ohrazda, 2004], it has to be said that the real teaching process is verycomplex, much more than described in this model. For instance, recent research and development streamseven focus on the applicability of workflow technology to realise a flexible e-learning model, as outlinedby [Lin et al., 2002]. Nevertheless, the didactical model in this subsection provides a formal specificationof an e-teaching process. Together with the content and the pedagogical model this formal descriptionmight be of use for developing and evaluating learning platforms as well as planning research activitieson technology-based learning and teaching. Further, it provides a theoretical basis for the next chapter,which is about adaptive e-learning.

3.5 Conclusions

Overall, it can be stated that traditional learning and teaching have a lot of similarities to e-learning, butalso differ in a few aspects. While learning paradigms, pedagogical aspects as well as didactical planningand course evaluation are rather equal for both, the classroom and the e-learning situation, implementingand running a course as well as the assessment of the learning process vary strongly. Assessing thelearning process and reacting to problematic aspects in learning are particularly difficult due to the factthat the learning and teaching process take place asynchronously.

Dealing with pedagogical and didactical issues of technology-based learning and teaching, this chapteroutlined that a lot of research is done in this field and still many aspects can be examined. Nevertheless,holding courses fully in a virtual way underlie certain disadvantages, wherefore various main researchstreams try to improve the e-learning process. One of these approaches deals with adaptive behaviour ine-learning platforms, as investigated in the next chapter on the basis of theoretical findings and the formalmodels of this and the last chapter.

Chapter 4

Adaptive E-Learning

“ At this stage, it is necessary to re-think the functions and limits of [e-learning]and try new modes of cognition in an innovative way. ”

[ Konrad Ginther, adapted by the author ]

Adaptive e-learning, as stated for example in [Shute and Towle, 2003], comprises a research and de-velopment stream dealing with educational systems that adapt the learning content as well as the userinterface with respect to pedagogical and didactical aspects. Adaptive educational hypermedia (as part ofadaptive hypermedia) is relatively new, having started around 1990 ago [Brusilovsky, 2004a]. Neverthe-less, the historical development of the basic principles and realised systems are often ignored as a resultof inconsistent definitions of terms as well as missing links between technical approaches and theoreticalaspects coming from other domains like pedagogy or didactics.

In particular, the problem concerning unfamiliar terms like programmed instruction, adaptive instruc-tional design, intelligent tutoring, etc. prohibits a holistic view of adaptive e-learning. [Park and Lee,2004] state that certain concepts of adapting instructions (e.g. Pressey’s adaptive teaching machine) canbe traced back to the early 20th century, while first systems like macro-adaptive instructional systemsor intelligent tutoring systems were realised in the 1960s. Moreover, [Corno and Snow, 1986] go backmuch further in history and outline that, “since at least the 4th century BC, adapting has been viewed as aprimary factor for the success of instruction”. Of course, these approaches were non-technological ones.

To give a holistic view of the field of adaptive e-learning, section 4.1 and section 4.2 examine historicalapproaches as well as system types in this area. Subsequently, section 4.3 determines basic terms andconcepts of adaptive e-learning based on the findings of the last two chapters, adaptation systems andtechnology-based learning and teaching. In section 4.4 the formal models of e-learning are extended byaspects of adaptation according to relevant issues identified in this chapter, before this chapter as well asthe theoretical part of this dissertation are concluded.

4.1 Historical streams

With respect to the theoretical models of adaptive e-learning, four main approaches which are used togive a historical overview can be identified: the macro-adaptive approach, the aptitude-treatment inter-action approach, the micro-adaptive approach and the constructivistic-collaborative approach. The firstthree approaches described and classified by [Park and Lee, 2004] are restricted to an old-fashioned viewof e-learning which is focused on the content and the learning process itself. With respect to youngerlearning paradigms and technologies, the last approach deals with topics like constructivism and adaptivecollaboration. All four approaches will be discussed closer in the following subsections.

47

48 4. Adaptive E-Learning

4.1.1 Macro-adaptive approach

First of all, the macro-adaptive approach which can be traced back to the 1970s is about adapting in-structions on a macro-level by allowing different alternatives in selecting a few main components suchas learning objectives, levels of detail, delivery system, etc. In this approach, instructional alternatives areselected mostly on the basis of the student’s learning goals, general abilities and achievement levels in thecurriculum structure. As adaptation decisions are determined before instruction – for example on the basisof rules – the macro-adaptive instructional approach can be characterised by the concept of adaptability.

[Corno and Snow, 1986] provide a taxonomy for systematic guidance in selecting instructionalmeditation – activating, modelling, participant modelling or short-circuiting knowledge – depending onlearning objectives – developing new skills or compensating students’ weaknesses – and student apti-tudes – intellectual abilities and prior achievement, cognitive and learning styles, academic motivationand personality.

[Glaser, 1977] introduces a more practice-oriented model for a macro-adaptive e-learning systemwhich supports defining preconditions for learning content, developing the appropriate competencies,adapting to the students’ learning styles and achieving different types of instructional objectives accord-ing to individual needs or abilities. On the other side, Glaser identified several conditions for a successfulimplementation of this approach, which partially explains, why macro-adaptive instructional systems havenot been as popular as hoped.

4.1.2 Aptitude-treatment interaction

The second approach comprises adaptation of instructional procedures and strategies to specific stu-dent’s characteristics. As suggested by [Cronbach, 1957], an e-learning environment serving a widerange of students requires a wide range of environments suited for optimal learning of the individual. Thisstrategy termed as “aptitude-treatment interaction” (ATI) proposes different types of instructions or evendifferent media types for different students. Several studies have been conducted to find linkages betweenlearning and aptitudes. The most important classes of learner characteristics can be summarised with thefollowing ones: intellectual abilities, cognitive styles, learning styles, prior knowledge, anxiety, achieve-ment motivation and self-efficiency (compare with the learner characteristics in section 3.2). Due to manystudies about measuring intellectual abilities, only a few experiences about the benefit for e-learning areresearched.

[Tobias, 1989] pointed out a number of difficulties for this approach like the dependency on the subjectarea, the poor applicability to actual classroom situations, growing abilities during learning process, etc.Therefore, he proposed an alternative model, the achievement-treatment interactions, to reduce some ofthe difficulties. This model focuses on task-specific variables relating to prior achievement and subject-matter issues. However, the fluctuating abilities and characteristics of the learner – a major problem ofthe ATI approach – still cannot be solved by idea of achievement-treatment interaction. Furthermore, thismodel also has the problem that useful information may be lost by not observing possible influences offactors like intellectual abilities, cognitive styles, anxiety and motivation.

Another important concept of adaptive instruction is learner control which deals with supporting thelearning process according to different abilities of the students by giving them full or partial control overthe style of the instruction or the way through the course content. Therefore, [Snow, 1980] defines threelevels of control: (1) complete independence, (2) partial control within a given task scenario and (3) fixedtasks with control of pace. Concerning learner control, it is proven that the success of different levels oflearner control is strongly dependent on the students’ aptitudes, for example it is better to limit the controlfor students with low-prior knowledge.

Despite the problems of the ATI approach pointed out here, faith in this approach is still alive and the

4.1. Historical streams 49

research is on-going. [Carrier and Jonassen, 1988] proposes an eight-step model to provide practical guid-ance for applying the ATI model to the design of courseware. According to this model, the course designerhas to identify objectives for the courseware, specify the tasks, define the relevant learner characteristicswith regard to the target group, determine how to adapt the instructions and design alternative treatments.As important learner characteristics influences the learning process, several methods like remedial, capi-talisation, preferential, compensatory and challenge for instructional adaptation are recommended.

[Park and Lee, 2004] states that this model seems to be the most practice-oriented within the ATIresearch, the other ATI approaches are considered to be very theoretical, problematic or time-consuming.Finally, it is also important to mention that an ATI-based system may not produce results differing fromnon-adaptive instructional systems without coherent and traceable rules to link the different learner andlearning variables to different tasks and instructional strategies.

4.1.3 Micro-adaptive approach

The third main approach to adaptive instructional learning is about adapting instructions on a micro-level by diagnosing the student’s specific learning needs during instruction and providing instructionalprescriptions for these needs. While the benefit of ATI research is either poor or not proven, severalstudies have shown that aptitude constructs are relevant to instructional and learning strategies.

Therefore, researchers developed micro-adaptive instructional models which use on-task instead ofpre-task measures. [Federico, 1983] states that monitoring the learner’s performance and behaviour, forexample by means of response latencies, response errors, emotional or motivational states, etc., can beexploited for manipulating and optimising instructional treatments and sequences on a much more refinedscale. The oldest model for the micro-adaptive approach is the idea of programmed instruction, originallyimplemented within a mechanical assessment device by [Pressey, 1926].

Adaptive e-learning in terms of the micro-adaptive approach is comparable to one-on-one tutoringand has to be separated in two main processes: The first part can be characterised as a diagnostic processassessing learner characteristics, such as aptitudes or prior knowledge and indices of the task, for instancedifficulty level, content structure or conceptual attributes [Rothen and Tennyson, 1978]. Referring to theprinciples of adaptation systems in section 2.3, this observation process comprises the concept of assessingand observing the adaptation information.

Based on the assessment of such on-task factors and states, the second part of micro-adaptive instruc-tion can be described as a prescriptive process optimising the interaction between the learner and thetask by automatically adapting the composition and sequencing of instructions according to the students’aptitudes and recent performance. Thus, it is necessary to define a strategy for selecting the appropriateamount of instruction and time to achieve a given learning objective. Examining this process within thescope of adaptation systems, the prescriptive rules would comprise the so-called adaptation procedures,while the systematically adaptation would be initiated on adaptation rules triggered by certain states in theadaptation information.

From the technological viewpoint, a number of micro-adaptive instructional models have been de-veloped. These models differ from programmed instruction in the way that they implement a particularmodel or theory of learning. Such a micro-adaptive model uses the time-dependent states of learner abil-ities and characteristics, especially the dynamically changing ones. As outlined by [Suppes et al., 1976],micro-adaptive instructional models often utilise a quantitative representation in order to determine andadjust learning contents during instruction in an accurate way. With respect to existing models, such asthe mathematical model, the trajectory model, the Bayesian model, the algorithmic approach and so forth,micro-adaptive instructional learning mainly deals with adapting few instructional variables, for examplethe amount of content to be presented or the presentation sequence of the content.

Another aspect of micro-adaptive instructional learning is response sensitivity, where computer-based


learning systems apply to simple student-computer interactions such as multiple-choice, short-answertypes, etc. AI-techniques for natural language dialogues provide an opportunity to apply the response-sensitive strategy, but are still limited and their development is relatively slow. As technology was notadvanced enough to implement response-sensitive diagnostic and prescriptive algorithms outside a lab-oratory until the late 1960s, the development of computers and new human-computer interfaces – likeeye-tracking devices – provides powerful tools for realising micro-adaptive learning systems. Currently,approaches like automated essay grading [Williams and Dreher, 2004] or natural language processing[Jurafsky and Martin, 2000] begin to bring along a certain degree of maturity and applicability.

Finally, interactive communication is an important element in adaptive instruction. The develop-ment of a powerful instructional system requires a communication model which considers the process ofinteractions between the student and the tutor. [Seidel et al., 1969] defines two channels for the learningprocess, the teaching channel to provide the content and the assessment channel to observe the learningprogress. Most student-system interactions in adaptive instruction are based on diagnoses of the students’interaction with the system.

4.1.4 Constructivistic-collaborative approach

During the eighties and early nineties, adaptive computer-based instructions focused mainly on the acqui-sition of conceptual knowledge and procedural skills [Andriessen and Sandberg, 1999]. Computer-basedlearning systems were criticised by many researchers for their limited ranges and adaptability of teach-ing actions compared to the large variety of strategies employed by a human expert teacher. In the latenineties, researchers began to examine approaches such as meta-cognitive strategies, collaborative andconstructivistic learning and motivational competence in adaptive instructional systems. The approachfocuses on modern aspects about how an e-learning system can be used within the learning process anddeals with pedagogical approaches like constructivism, the Vygotsky’s Zone of Proximal Developmentand Contingent Teaching, etc. An important element of this approach is the usage of collaborative tech-nologies which are considered often to be an essential component of e-learning, as stated for example by[Lennon and Maurer, 2003].

By the means of the constructivist learning paradigm (see also section 3.1), the learner plays anactive role in the learning process constructing his own knowledge through experience in a context inwhich the target domain is integrated. [Akhras and Self, 2000] argues that constructivistic learning maybenefit from a system’s intelligence including mechanism of knowledge representation, reasoning anddecision making. An adaptive system enables learning by focusing on how knowledge is learned andhas the following components: context, activity, cognitive structure and time extension. The contextshould be flexible enough to support different levels of learning experiences. The cognitive structureshould be designed carefully, so that previously constructed knowledge influences the way to interpretnew experiences.

According to [Vygotsky, 1978, p. 86], providing immediate and appropriately challenging activitiesand contingent teaching based on learner’s behaviour is necessary for them to progress to the next level.Besides, Vygotsky’s zone of proximal development implies that a minimal level of guidance is bestfor learners. There should be no user model of a learner, because the learner’s performance is local andsituation constrained by the current activity. However, these contingent-based learning is limited in termsof the own problem-solving strategy learners develop and the problem of diagnosing such a complexmodel.

Some younger adaptive e-learning approaches take account of students’ motivational factors com-bining the instructional plan with a “motivational” plan. As outlined out by [Wasson, 1990], instructionalplanning can be divided into two streams: Content planning for selecting the next topic to teach and adelivery planning for determining how to teach the selected topic. Motivational components should be

4.2. Types of adaptive educational systems 51

considered while designing delivery planning.

Meta-cognitive skills describe students’ understanding of their own cognitive process, which affectsthe learning process itself, according to psychologists like Dewey, Piaget and Vygotsky. For instance,[White et al., 1999] states that meta-cognitive processes can be easily understood and designed a con-ceptual framework integrating cognitive and social aspects of cognition by strategic advice and guidancewithin the learners’ research projects.

A new pedagogical approach to adaptive instructional systems is to support collaborative learning ac-tivities which can be a powerful learning experience as proven by studies. Referring to [Soller, 2001],the following characteristics of effective collaborative learning can be identified: (1) participation, (2)social behaviour, (3) performance analysis, (4) group processing and conversation skills and (5) primitiveinteraction. Based on these characteristics, components for a collaborative learning system, such as acollaborative learning skill coach, an instructional planner, a student or group model, a learning compan-ion and a personal learning assistant, can be derived. Using such components, online courses could beextended from instructional design for individual learners to collaborative activities for groups of learners.

4.2 Types of adaptive educational systems

According to these four historical streams for adaptive instructional learning, this section treats e-learningsystems based on one or more of these theoretical approaches. First, exemplary systems for the macro-adaptive approach are shown. Second, computer-managed instructional systems, intelligent tutoring sys-tems and adaptive hypermedia are examined. Finally, components considering modern pedagogical as-pects and collaborative elements are pointed out.

4.2.1 Macro-adaptive instructional systems

In the early 20th century, adaptive e-learning followed the basic ideas of the macro-adaptive approachby which students were grouped or tracked by tests scores. In this period a couple of adaptive instruc-tional systems were developed in order to support different learner abilities. For instance, [Reiser, 1987]describes projects like the Burke plan or the Dalton plan, where students were allowed to go throughthe materials at their own pace. Additionally, these approaches included elements like presenting or ex-plaining specific learning content, asking questions in order to monitor the learning progress and provideappropriate feedback for the students.

In 1963, a macro-adaptive system was developed at Columbia University [Keller, 1968]. This system,namely the Keller plan, provides personalisation for each student and offers features like required mas-tery to proceed to the next unit, usage of textbooks and workbooks, etc. [Flanagan et al., 1975] reporton the realisation of the Program for Learning in Accordance with Needs (PLAN), a system to supportstudents with options for selecting appropriate learning objectives and materials. In the early 1970s, thisprogramme was conducted in more than 100 elementary schools.

In 1965, a more comprehensive macro-adaptive e-learning system, namely the Individually GuidedEducation (IGE), was developed at the University of Wisconsin [Klausmeier, 1976]. In IGE, instructionalobjectives are first determined for each student, for example on the basis of academically profile includ-ing previous achievements or diagnostic assessments. This information allows the teacher to determinenecessary guidance and select alternative materials, for example text, audiovisuals or collaborative tasks.

According to [Bolvin and Glaser, 1968], the Individually Prescribed Instructional System (IPI) whichwas developed at the University of Pittsburgh in 1964 provides learners with adaptive instructional en-vironments. In the IPI, diagnoses are made before, during and after a unit to adapt the instruction toprior knowledge and learning objectives and to determine the student’s mastery. Extending the IPI with


more and other diagnosis methods and instructional prescriptions, the Adaptive Learning EnvironmentsModel (ALEM) was developed, as pointed out in [Wang, 1980]. This system provides extended featureslike a powerful instructional management, guidance for involving parents in learning activities at home,procedures for team teaching and group activities, etc.

4.2.2 Computer-managed instructions

Another type of system originated on the basis of the macro-adaptive approach is the class of ComputerManaged Instructional (CMI) systems [Park and Lee, 2004], which is related to Computer Assisted In-struction (CAI) [Maurer, 1985]. CMI systems offer functions for diagnosing learning needs and prescrib-ing instructional activities appropriate for these needs. As an example, the Plato Learning Management(PLM) system provides tests on different instructional levels, such as a module, a lesson, a course and acurriculum. According to the performance of a student, specific instructional prescriptions like repeatingthe assessment or the whole unit, offering additional instructions for a course, etc. are provided. Whenmastery of all objectives in the module has been reached, a student may proceed to the next module.

As pointed out here, CMI systems provide many important macro-adaptive instructional features al-lowing a teacher to monitor and control the student’s learning. However, the increase of personal computercapabilities enables aspects of personalisation within CMI systems. [Ross and Morrision, 1988] describethe development of a CMI system implementing features of macro and micro-adaptive models. Contraryto other macro-adaptive instructional systems and programs, CMI systems are much more effective interms of adaptability of the online learning process.

4.2.3 Intelligent tutoring systems

Intelligent Tutoring Systems, abbreviated as ITS, can be defined as adaptive instructional systems applyingAI techniques. [Shute and Psotka, 1996] mention that ITSes are developed to resemble the process of theone-on-one learning process between teacher and student. ITSes have to represent the content, implementthe instructional strategy and provide mechanism for assessing the student’s learning progress. Thesefeatures can be summarised with the following components: the student-modelling module for assessingthe student’s current state and determining his conceptions and reasoning strategies, the expertise modulefor generating instructional contents on the basis of the learner’s performance and the tutoring module forselecting and presenting instructional material.

[Tennyson and Christensen, 1988] proposes a two-level model of adaptive instruction combiningmicro-adaptive instructions and aptitude variables: First of all, this computer-based model allows theexpertise module to establish conditions of instruction based on the learner’s characteristics. Secondly,the tutoring module provides moment-to-moment adjustment of instructional conditions by sequence ofinstruction, adapting the amount of information, display time, example formats, etc. The micro-leveladaptation takes place on the learner’s on-task performance. The procedure itself can be considered asresponse sensitive.

AI methods can be used for the representation of knowledge or natural language dialogues to adaptthe contents to the student and allow a more flexible interaction between the system and the student. ITStechniques provide powerful tools for effectively capturing the learning and teaching process, supportingmeta-cognitive processes, etc. However, it has been criticised that developers have failed to considerimportant learning principles and instructional paradigms, as outlined for instance in section 3.1. Themost challenging problems comprise issues about how to learn and teach with technologies like artificialintelligence.

4.2. Types of adaptive educational systems 53

4.2.4 Adaptive hypermedia

Starting about 1990, the field of Adaptive Hypermedia (AH) which is inspired by intelligent tutoringsystems arose. Adaptive Hypermedia Systems (AHS) try to combine hypermedia-based and adaptive in-structional systems, where adaptive and personalising interfaces were integrated into hypermedia systems.Functional aspects of AHS mean components that may not be visibly, such as an altering behaviour of the“next” button of the interface. An AHS should “be based on hypertext link principles, have a domainmodel and be capable of modifying some visible or functional parts of the system on basis of the informa-tion contained in the user model” [Eklund and Sinclair, 2000]. AHSes have been employed for educationalsystems, e-commerce applications, online information systems and online help systems. Because of itspopularity and accessibility, most adaptive educational systems have focussed on the Internet since 1996.

Adaptive hypermedia methods can mainly be divided into two areas of adaptation, the content-leveladaptation or adaptive presentation, where the content is assembled or presented in different ways or orders[DeBra, 2000] and the link-level adaptation or adaptive navigation support, where links are generatedaccording to different methods like direct guidance, adaptive sorting, adaptive annotation and link hiding,disabling and removal [Brusilovsky, 2000]. As an example of direct guidance, the system ELM-ARTgenerates additionally dynamic links to connect to the next most relevant node to visit. Contrary to this,the HYPERTUTOR system hides links which are not relevant to the user’s current task. InterBook andAHM are further examples of hypermedia systems applying the annotation technique, where links arenamed according to the user’s knowledge.

Adaptive hypermedia systems include models for a user’s goals or tasks, domain and backgroundknowledge, preferences, etc. Consequently, these models are utilised for adaptation decisions. Further-more, it is possible to assess the long-term interests as well as the short-term goals of users [Fink et al.,1998]. In systems that consider these aspects, the user’s interests serve as a basis for recommending rel-evant content. Moreover, individual traits, such as cognitive factors, personality and learning styles areof importance, although researchers disagree about which characteristics can and should be used. Finally,adaptation to the user’s environment is a new kind of adaptation fostered by web-based systems which hasbecome an important issue due to the usage of different hardware, software and platforms.

The introduction of hypermedia has had a great impact on adaptive instructional systems. While otherkinds of adaptive systems cannot be realised without programming skills, the adaptive courses for AHScan be created with recent authoring tools, for example SmexWeb. However, there are limitations to AHS:Often they are theoretically or empirically not well founded. In particular, the evidence of effectivenessof AHS is shown only for some few aspects [Specht and Oppermann, 1998]. Moreover, [DeBra, 2000]outlines that missing or omitted prerequisite relationships in AHS may guide the user to pages not relevantor not understandable for them. Further, assessing learners’ knowledge states is the most critical factor fora successful implementation of an AHS.

4.2.5 Other technologies in the field of adaptive e-learning

As new pedagogical approaches and technologies came up, adaptive e-learning was extended by innova-tive systems which will be discussed within this subsection. First of all, the paradigm of constructivis-tic learning brought up systems like Intelligent Constructivistic Environment for Software Engineeringlearning (INCENSE) described by [Akhras and Self, 2000]. INCENSE offers features to analyse a time-extended process of interaction between the learner and a set of software-engineering situations and toprovide a learning situation based on the learner’s goals to support further processes of learning experi-ences rather than acquisition of target knowledge. As described by [Dietinger et al., 1999] and [Garcia-Barrios et al., 2002], a Dynamic Background Library can be used in terms of constructivistic learning toprovide dynamically retrieved and up-to-date knowledge managed by experts.


[Wood, 2001] reports on examples of tutoring systems based on Vygotsky’s zone of proximal devel-opment. For example, ECOLAB, which helps children aged 10 to 11 years learn about food chains andwebs, provides challenging activities and the right quantity of assistance. Other examples of Vygotsky’sapproach are QUADRARIC, DATA and EXPLAIN. [Gredler, 2004] gives an overview of games and sim-ulations, which can be used to mediate a model to the learner or provide a journey through a domain on aplayful way. Adaptation can be realised by different levels of complexity, levels of speed, or even tutoringcomponents. As examples like Underwater SeaQuest or SimCity points out, these kinds of elements arenot only applicable to children, but also to adults. Additionally, [Kuhn and Gudjonsdottir, 1999] reportabout a virtual campus approach by means of a 3D multimedia educational environment named ViKar,which improves online learning through providing a virtual reality with multi-user communication andcollaboration.

Systems considering the motivational state of the learner try to incorporate gaze, gesture, nonverbalfeedback, etc. to detect and increase students’ motivation. For example, COSMO includes a pedagogicalagent that can adapt its facial expression, its tone of voice, its gesture and the structure of its utterancesduring its interactions with learners. Another system named MORE can detect the student’s motivationalstate and reacts to motivate the distracted, less confident or discontented student [DuBoulay and Luckin,2001]. An example of focusing and improving meta-cognitive skills is the Geometry Explanation Tutorprogramme [Aleven et al., 2001]. As an effective meta-cognitive strategy, this system explains examplesor problem-solving steps to help students learn with greater understanding. The tutor is even able torespond to incomplete statements in the student’s explanations.

Finally, [Park and Lee, 2004] report on systems implementing adaptive collaborative e-learning.Such systems can be classified in terms of their application as follows:

• Computer-based collaborative tasks (CBTC) like the Envisioning Machines support group learningand group activity by presenting a task for the group and providing collaboration via intelligentcoaching.

• Co-operative tools (CT) like the Case-Based Reasoning Tool or the Writing Partner describe sys-tems taking over some of the burden of lower-order tasks, while students work with higher-orderactivities.

• Furthermore, intelligent co-operative systems (ICS) like DSA, PeoplePower or the Integration-Kidsystem can be seen as an intelligent co-operative partner, a co-learner or a learning companion.

• Computer-supported collaborative learning systems (CSCL) serve as a communication interfacesuch as a chat tool or a discussion group supporting collaboration between learners. Such systemsprovide the least adaptability to learners.

Although these four kinds of collaborative systems are in an early developmental stage, they can beconsidered as important aspects of adaptive e-learning, because they do not only facilitate group activities,but also help educators gain further understanding of group activities and determine collaborative tools forlearning.

4.3 Existing theoretical models for adaptive e-learning

As adaptive e-learning has a rather long history, there already exist many theoretical frameworks and evenformal approaches. In this section the main categories for these models are examined by pointing out themost prominent examples of each category.

4.3. Existing theoretical models for adaptive e-learning 55

4.3.1 Informal frameworks

The first category deals with informal models for adaptive e-learning. Amongst them, [Park et al., 1987]describe a powerful theoretical framework considering aspects of micro and macro-adaptive instruction onthe basis of learner characteristics. As shown in figure 4.1, this conceptual model presents necessary func-tional entities for the adaptation process in the field of learning. Macro-adaptive instructional behaviour isimplemented by the outer feedback cycle between input and output, while micro-adaptation of the learn-ing process is realised within transactions, based on diagnostic and tutorial rules to adapt the knowledgeand instructional presentation. Referring to adaptation systems, this approach can even be characterisedto be meta-adaptive, because the refinement of tutorial rules is considered.

Figure 4.1: Model of adaptive instruction, adopted from [Park et al., 1987]


Furthermore, [Shute and Towle, 2003] comprise a framework for two types of assessment. As shownin figure 4.2, adaptation may be related to the domain-dependent and domain-independent learner model.Therefore, this approach differentiates between adaptation and instruction, while adaptation rules aimto select an appropriate learning object and instructional rules deal with adapting the course structure.Analysing this models, to some extent analogies to section 3.4 (content, pedagogical and didactical model)can be found, although the last chapter did not deal with adaptation in e-learning environments so far.

Figure 4.2: Framework for adaptive e-learning, adopted from [Shute and Towle, 2003]

4.3. Existing theoretical models for adaptive e-learning 57

Addressing technological frameworks, [Brusilovsky, 2004b] introduces a generic and flexible archi-tecture, the so-called KnowledgeTree (see also figure 4.3). This technological approach aims at providingadaptive e-learning functionality by means of specialised services adapting various aspects of the learningcontent or the user interface. At the back-end, the student modelling server CUMULATE manages theprofiles and models of the learners, while the front-end consists of a portal, for example the Knowledge-Sea. Between these two layers, activity servers and value-adding services implement aspects of adaptivityor personalisation. For example, such activity services could be WebEx which serves interactive andannotated programming examples or QuizPACK which provides parameterised programming questions.

Figure 4.3: The KnowledgeTree architecture, adopted from [Brusilovsky, 2004b]

These two informal models for adaptive e-learning already demonstrate that theoretical frameworkscan be of significant importance for designing or evaluating adaptive behaviour in instructional systems.Generally, both of them fulfil requirements given from adaptation systems, e-learning and adaptive instruc-tional approaches. Nevertheless, despite such visually and verbally described models, other researcherstried to formalise adaptive behaviour within the field of e-learning as shown in the next sections.

4.3.2 Knowledge- and workflow-based approaches

A more formal generation of theoretical frameworks for adaptive e-learning focuses on learning contentor activities. On the one side, cognitive psychologists examine learning on the basis of skills, relationsbetween them and learners’ knowledge states. For instance, [Albert and Hockemeyer, 1997] highlighthypertext-based learning with respect to the so-called Knowledge Space Theory. Applying a rathermathematical approach, this model suggests considering relations between such skills and personalisingthe learning path for students according to this information. Later, [Albert and Hockemeyer, 2002] reporton applying the knowledge space theory for adapting the course structure, for adaptive assessment of thelearners’ knowledge state and for adaptive training.

On the other side, e-learning can also be seen as a sequence of learning objects. As one example ofsuch an approach [Karampiperis and Sampson, 2005] introduce an adaptation model proposing the bestsequence of learning objects through a course. Therefore, all possible sequences are calculated on the basisof concept selection and the knowledge space and the best fitting learning path is selected according to thelearner characteristics, whereby the suitability of each sequence is calculated along ratings for learningobjects given by instructional designers.


Inspecting both these models, it can be stated that these approaches focus mainly on the sequencing oflearning content. [Lin et al., 2002] even compare the learning process to workflows of learning activitiesand, therefore, utilise workflow management technologies. Nevertheless, these models lack aspects likeadapting the user interface, aggregating or visualising instructions in different ways or adding instructionalunits on demand.

4.3.3 Formal and logic-based models

The application of formal approaches is common in other areas, for example in the field of adaptivesystems as shown by [DeJong, 1975, p. 5ff]. In the field of adaptive e-learning, a few researchers alsotried to describe the systemic behaviour of e-learning environments providing adaptivity and adaptabilityin some way. Two of them are highlighted in this subsection.

The Adaptive Hypermedia Application Model (AHAM) by [DeBra et al., 1999b] is based on theDexter model and, therefore, focuses on hypertext and hypermedia systems. AHAM allows specifyingrelevant didactical issues as well as pedagogical rules. Yet, this formal model restricts the possibilities ofteachers, for example by merely addressing the cognitive domain (read, learned) or not separating betweenlearning materials and learning activities. On the other side, pedagogical aspects are linked to attributes ofthe user-model, so that it is not possible to define pedagogical rules depending on the context or domainof a course. Finally, AHAM deals with architectural issues to a great extent, which is not a requirementfor designing and evaluating adaptive behaviour within an e-learning environment.

Another very powerful formal approach to characterisation of adaptive educational hypermedia isthe one by [Henze and Nejdl, 2004]. Contrary to other models, this one comprises a first-order logicrather than some theoretical framework for a component-based architecture or for a certain adaptationmethod. Thus, it is possible to evaluate and compare adaptive e-learning environments with respect totheir behaviour instead of their architecture or features. In detail, the logic-based model by Henze andNejdl are based on the definition of a document space, a user model, observations and an adaptationcomponent. Further, different predicates – like “part-of”, “prerequisite”, “is-a”, “is-dependent”, “has-property”, etc. – are used to build up the first three sub-models. Subsequently, the adaptation componentconsists of rules describing how adaptation is performed within a system.

In accordance with section 2.4, this first-order logic can be seen as the most powerful way to describeadaptive behaviour of systems. Yet, in the field of adaptive e-learning practitioners might be lost havingonly guidance through some exemplary systems in [Henze and Nejdl, 2004]. Although it is possible todescribe all pedagogical and didactical issues – as given in the previous chapter for example – it wouldbe helpful to have a conceptual framework for adaptive e-learning, which is attempted by section 3.4(content, pedagogical and didactical model) and the next section (adaptation model).

4.3.4 Other theoretical frameworks

Another framework realised with UML is the so-called Munich Reference Model introduced by [DeKoch,2000, p. 73ff]. Again, this model comprises adaptive hypermedia applications based on the Dexter Hy-pertext Reference Model [Halasz and Schwartz, 1994] from a more component-based viewpoint. There-fore, its applicability for designing and evaluating adaptive e-learning environments from the behaviouralviewpoint is rather restricted, and it can be considered to be the theoretical background to the SmexWeb[DeKoch, 2000, p. 287ff].

Furthermore, [Tochtermann and Dittrich, 1996] introduces the Dortmund Family of Hypermedia Mod-els (DFHM), another formal approach to adaptive hypermedia which is even based on VDM-SL. Neverthe-less, this model deals rather with an object-oriented, component-based architecture than with a systemicbehaviour and, further, addresses primarily hypermedia, not educational hypermedia.

4.4. Formalising adaptive behaviour in e-learning systems 59

Younger streams, such as AEHS or personalised e-learning services as pointed out in [Brusilovsky andNijhavan, 2002] or [Conlan, 2005], tie up to the theoretical foundations of ITS and comprise the followingmodels: (a) a content model semantically enriching the learning content, (b) a learner model describinglearner characteristics, (c) an instructional (or tutoring) model providing the teaching strategy and (d) anadaptation model including the possibilities and rules to adapt the learning process.

Notwithstanding the existence of different formal models – a few others are analysed in [DeKoch,2000, p. 62ff] – the next section attempts to build up another formal approach to provide a flexible wayfor defining, describing and evaluating adaptation of instructions and the platform’s elements on the basisof relevant aspects of e-learning.

4.4 Formalising adaptive behaviour in e-learning systems

In accordance with [Henze and Nejdl, 2004], a formal model for adaptive e-learning is useful for designingand evaluating adaptive e-learning environments. As adaptive e-learning deals with adaptation systems aswell as with technology-based learning and teaching, this section combined aspects of chapter 2 andchapter 3 in order to build up such a theoretical framework.

4.4.1 Influences of adaptation systems

Concluding from chapter 2, adaptation systems play an important role in the field of adaptive e-learningenvironments. Firstly, adaptation of the learning process can be characterised along the concepts of adapt-ability, adaptivity, customisation and personalisation. As an example, adaptability would comprise theinstructor’s adapting the learning process, while a learner could customise parts of the user interface forexample.

Further, automatically adapting course content or the user interface on the basis of some didacticalmodel, for example a knowledge-based model, is considered to be adaptive, if the adaptation processis independent of any learner characteristics. Personalised e-learning would include that adaptation iseffected by at least one learner characteristics. In the context of this dissertation, an “adaptive e-learningenvironment” can be defined as a learning management system satisfying the requirement that it providesat least one adaptive feature.

Secondly, adaptive e-learning requires certain models which have to be observed in order to triggerthe adaptation process:

• The adaptation information could, for instance, deal with learner characteristics, for example certainknowledge states or learning styles.

• Adaptation rules could comprise certain achievement levels or thresholds for learner characteristics,like the determined learning objectives or a factor for a learning style.

• The adaptation procedures prescribes, which parts of the user interface and the learning content –no matter whether visible or not – are adapted in which way.

• Adaptation targets often aim at the learning outcome by means of the result of formative and sum-mative assessment. Yet, it is also thinkable to define certain behaviour for some factor of the learningprocess, for example the learner’s level of attention.

Thirdly, a meta-adaptive strategy for the adaptation of the learning process is recommended due toimproving the adaptation strategy and the knowledge transfer. Yet, the development and empirical eval-uation of a meta-adaptive e-learning strategy might be much harder than of the adaptive behaviour. As


a consequence, adaptive e-learning environments can mainly be classified as deterministic or heuristicsystems. In the case of meta-adaptation, such a system might also be a purpose-seeking one.

From the viewpoint of systems theory, adaptive e-learning systems can be seen as second-order cyber-netics consisting of an observable part (user interface and presentation layer) and an observer (adaptivecomponent). Addressing systems thinking, an adaptive e-learning environment can be characterised as in-telligent, if it implements meta-adaptive behaviour. Finally, it has to be stated that the adaptive behaviourof an e-learning system can be described utilising a formal specification language like VDM-SL.

4.4.2 Influences of technology-based learning and teaching

Addressing aspects of e-learning and e-teaching examined in chapter 3, the following conclusions canbe drawn for the field of adaptive e-learning: First of all, two adaptation processes can be identified, adidactical one and a pedagogical one. While didactical adaptation deals with domain-specific aspectslike the learner’s pre-knowledge or knowledge states only, pedagogical adaptation comprises the idea ofadapting the learning content according to certain learner characteristics, for example by means of thelearning styles or intellectual capabilities.

Secondly, didactical adaptation of the online learning process – which seems to be more importantfrom the viewpoint of a teacher – are closely related to e-teaching dealing with didactical input and out-put, for example on the basis of the determined learning objectives and assessment of the learning process.In this context, researchers in the field of adaptive e-learning are interested in the following issues: (1)dependency between learning objectives and instructional design, (2) effective assessment methods fordifferent competency types and (3) creating and exploiting domain models in order to improve learn-ing. Overall, e-didactics aims at optimising the instructional sequence, so that learners achieve the givenlearning objectives in the most effective way.

Thirdly, the adaptation of learning with respect to pedagogical aspects is about delivering appropriateinstruction to the learner and adapting the user interface accordingly. Thus, this adaptation process can becharacterised as personalised content delivery and presentation. Research comprises the field of aptitude-treatment interaction, for example the examination of certain learning characteristics such as cognitive orlearning styles, as well as the influences of such factors on instructional design and learning. Further, theassessment of learner characteristics as well as didactical issues by using new methods or technologies iscurrently a favourite subject of research.

Concluding these influences of technology-based learning and teaching, adaptive e-learning can beunderstood as an attempt to implement didactical competencies, because teachers also have to assess thelearning process and adapt it to certain factors or learner characteristics. Therefore beside fundamen-tal mastery of subject matter, teachers have to apply different skills, for example classroom assessmentmethods, in order to improve the knowledge transfer within the classroom situation. As the realisationof online courses differs in the way that the learning process and the teaching process take place asyn-chronously, adaptive e-learning is one possible way to compensate the problem that the teacher cannotinfluence learning directly and immediately.

4.4.3 Model for adaptation of the learning process

Formalising adaptive e-learning requires, as already shown in the last chapter and the inspection of existingtheoretical approaches, the following models: (1) a content model, (2) a pedagogical model and (3) adidactical model can be described. These three models were already introduced in section 3.4. As adaptivee-learning was defined as an attempt to implement didactical skills, adaptive behaviour of a learningmanagement system can be described by simply extending the didactical model with respect to possibleadaptation methods, which is done for example in figure 4.4. The three models of the last chapter together

4.4. Formalising adaptive behaviour in e-learning systems 61

with the one in this section are, for reference reasons, subsumed under the name FORMABLE, which isthe abbreviation for “FORmal Model for Adaptive Behaviour in e-Learning Environments”.

Figure 4.4: Formal specification of the adaptation model

In this formal model the following aspects have to be outlined here:

• First of all, this approach is based on the fact that two adaptation processes exist, a didactical anda pedagogical one. From the viewpoint of knowledge transfer, the didactical one is of primaryimportance, because it includes didactical planning, instructional design and assessment.

• Second, FORMABLE is aware of three kinds of adaptation categories: (1) adapting instructions,(2) adapting instructional sequences and (3) inserting additional instructions. These three types ofoperation are sufficient to realise different kinds of adaptation methods as shown in the practicalpart of this dissertation. Adaptation of instructional level requires that there are alternative instruc-tions, which are relevant to a certain didactical objective or suitable for a certain pedagogical state.


Aggregation of instructions is only supported by fully exchanging an instruction with another onedue to the fact that knowledge artefacts are atomic entities. Adapting the instructional sequencecan be characterised in the way that, beginning with a certain instruction, the sequence of the re-maining instructions is re-ordered with respect to didactical dependencies. Finally, the insertion ofa new instruction also requires that its relevance for the course’s learning objectives as well as thepedagogical suitability is already given.

• Third, these three main operations for adaptation in e-learning environments do not so far differ-entiate between adaptability and adaptivity. Real adaptive behaviour would require some thread-mechanism assessing didactical and pedagogical states and the triggering of one of these threeoperations (as compared with the formal model of a multi-purpose adaptive system).

• Finally, FORMABLE does not consider whether adaptation takes place before or within instruction.Yet, it is possible to determine which learner-dependent or learner independent states are assessedand which elements of the learning process – parts of the user interface or the learning content – areadapted.

Overall, FORMABLE serves as a theoretical basis for adaptive e-learning and allows designing andevaluating methods, features and environments for adaptive e-learning, as shown in the practical part ofthis dissertation.

4.5 Conclusions

Summarising the theoretical part of this work, adaptive e-learning is considered to be related to two im-portant areas: Adaptation systems as well as technology-based learning and teaching. On the one side,systems theory and adaptation systems provide the basics, how adaptation in learning environments works,which components are necessary and how adaptivity can be realised. From a systemic viewpoint, this re-search field can be of importance for defining adaptive behaviour of e-learning system, for example byfocussing the designer’s attention to relevant systemic characteristics or special development streams ofsystems theory.

On the other side, e-learning and e-teaching deal primarily with research questions about how instruc-tional design and assessment should be determined according to learning objectives, which factors andlearner characteristics should be assessed and how the learning process can be adapted in an appropriateway. Pedagogy and didactics must particularly answer whether the implementation of adaptation methodsfor e-learning has a positive effect on the students’ learning and, therefore, is profitable by means of effortand results.

As pointed out in this chapter, adaptive e-learning deals a lot with compensating disadvantages ofonline learning, i.e. that the teacher cannot assess and adapt learning in real time. Based on an extensiveliterature survey, the theoretical model in this work, as with other frameworks and approaches, attempts todescribe adaptive behaviour within e-learning environments in order to support the design and evaluationof such systems. The applicability of FORMABLE is demonstrated in the practical part of this work,for instance in the next chapter which examines requirements for standards and specifications to supportadaptive e-learning.

Drawing a final conclusion from the theoretical part, it has to be stated that adaptive e-learning can begenerally seen as one possible approach to implementing pedagogical competencies of didactical expertswithin information technology.

II. Practical Aspects

63

Chapter 5

Towards Standardising Adaptable Cour-seware

“ The nice thing about standards is that there are so many of them to choose from. ”

[ Andrew S. Tanenbaum ]

As the theoretical issues given in the last three chapters were rather abstract, the second part of this dis-sertation aims to examine practical aspects from theory in order to show how the formal model of adaptivee-learning can be applied in practice. Therefore, this and the next two chapters deal with more pragmatictopics. This chapter examines aspects of standardised adaptable courseware, the next one attempts to de-scribe an ideal adaptive e-learning environment and the final chapter of the practical part summarises thetechnical realisation of the AdeLE prototype [AdeLE, 2006], a system which resulted from one researchproject in the field of adaptive e-learning.

Addressing adaptation within the online learning process, the FORMABLE model states that twomajor components of online courses can be adapted. The first one comprises the learning managementsystem, by means of varying learning activities as well as adapting the user interface. Considering an idealadaptive e-learning environment, such issues are examined in the next chapter. The second component fo-cuses on adaptation of the learning content, which is particularly relevant to didactical considerations.Standardisation might be particularly problematic in the context of adaptive e-learning due to the asym-metric nature of the online learning [Jain et al., 2002, p. xxvii].

Therefore, this chapter concentrates on aspects of standardising adaptable courseware. Beginningwith section 5.1, a short overview of the ongoing standardisation efforts and a few selected standards andspecifications in the field of e-learning is given. Thereafter, section 5.2 derives requirements for such anideal standard from the theoretical part of this dissertation. Furthermore, in section 5.3 restrictions ofcurrent specifications are pointed out, before section 5.4 introduces a standard-based approach to adaptivee-learning utilising a commonly-known set of specifications, namely SCORM.

5.1 Standardisation of learning content

Standardisation is targeted in many areas, such as for digital libraries [Paepcke et al., 1998], workflowmanagement [VanDerAalst, 1998], museum information systems [Moen, 1998], the World Wide Web[Berners-Lee, 1996] or e-learning [Duval, 2001]. Due to the necessity of high-quality courseware, in-teroperability issues like transferability and reusability of content as well as the usage of learning objectrepositories have to be considered [Qu and Nejdl, 2002]. Therefore, the standardisation process in the

65

66 5. Towards Standardising Adaptable Courseware

field of e-learning has lasted about two decades, but is still in progress and misses different requirements– particularly concerning adaptive e-learning – as pointed out later in this chapter.

5.1.1 The past

Referring to [Dietinger, 2003, p. 51], the oldest standardisation consortium in the area of e-learning is theAviation Industry CBT Committee [AICC, 2007b] examining and developing specifications for describingaspects of online courses since 1988. According to [AICC, 2007a], the AICC subcommittees aim tocreate guidelines and specifications for the following seven issues: Computer Managed Instruction (seealso section 4.2), Communication, Digital Electronic Library Systems (DELS), Independent Test Lab,Management and Processes, Training Infrastructure as well as Training Technology.

Beside the Dublin Core Metadata Initiative [DCMI, 2007] which focuses on adopting and developingspecialised metadata, the IEEE Learning Technology Standards Committee [IEEE, 2007] can be particu-larly outlined as one of the pioneers of the standardisation process for learning content. One commonly-known output of this approach is the so-called Learning Objects Metadata (LOM) standard allowing thedescription of instructional resources on the basis of nine categories [Duval, 2001]. For adaptation of theonline learning process particular the educational section of the LOM standard is of interest, as shown inchapter 7 of this work.

Thirdly, the Instructional Management Systems (IMS) Global Learning Consortium [IMS, 2007i] ini-tiated in 2001 has a deep impact on standardisation of courseware. Starting with a few specifications only,[IMS, 2007h] nowadays provides a set of 17 specifications, including all relevant aspects of e-learning,for example accessibility, competency, digital repositories, e-portfolios, resource list and much more. Be-sides, IMS also examines abstract frameworks for learning platforms, toolkits or vocabularies in the fieldof technology-based learning and teaching. Although many scientific institutions participate in the effortsof this consortium, IMS follows rather pragmatic principles and contributed several specifications to otherprojects and initiatives, for example the SCORM specification set.

With respect to [ADL, 2007a], the Sharable Content Object Reference Model (SCORM) by AdvancedDistributed Learning (ADL) is meant to be a “collection of standards and specifications adapted frommultiple sources to provide a comprehensive suite of e-learning capabilities that enable interoperability,accessibility and reusability of Web-based learning content”. As mentioned before, SCORM includessome specifications of the IMS Global Learning Consortium, but also for example the LOM standard.Overall, ADL tries to exploit the results of various consortia and projects in order to build up a practicalset of specifications to describe courseware. Moreover, ADL also provides a sample implementation ofa SCORM-compliant learning management system [ADL, 2007b]. As SCORM is utilised also in thepractical part of this dissertation, it is going to be described closer later on.

Finally, different projects – mainly research projects – also had a significant impact on standardisationof learning content. For instance, the “Alliance of Remote Instructional Authoring and Distribution Net-works for Europe” (ARIADNE, [ARIADNE, 2004]) founded by the European Community was mainly in-volved in the development of the LOM standard mentioned above [Duval et al., 2001]. Further, [Dietinger,2003, p. 68ff] reports on other projects like “PROmoting Multimedia Access to Education and Training inEUropean Society” (PROMETEUS) and initiatives like the “Schools Interoperability Framework” (SIF)in this area. Additionally, also companies like Microsoft investigated in this area and still are involved, asshown with the “Learning Resource iNterchange” (LRN) reference implementation [Microsoft, 2000].

5.1.2 The present

In fact, the standardisation process in the field of e-learning is still in progress and only a few specificationsare really standardised by an international organisation like the ANSI, as stated by [Gries, 2003]. Figure

5.1. Standardisation of learning content 67

5.1 visualises these organisations and projects from the past and the present as well as their relation tocertain specification and standards. While committees and projects like AICC, IMS, DCMI, ARIADNE,PROMETEUS, etc. deliver contributions to the standardisation process, other organisations like ADL orthe Open Knowledge Initiative [OKI, 2007] utilise this input and develop reference models for practicalapplication.

Figure 5.1: The development process of e-learning standards, adopted from [Gries, 2003]

Overall, only national organisations like the IEEE or the European Committee for Standardization[CEN, 2007] is able to submit specifications to a standardisation organisation like the American NationalStandards Institute [ANSI, 2007]. As a result, only a few real standards currently exist. The LOM metadataset is one of these few standards.

5.1.3 The future

The future of the standardisation efforts in the scope of e-learning is not predictable at all, particularlyas many research projects as well as commercial products focus on this topic. Some speculation in thecontext of this dissertation can be made anyhow:

• First of all, available standards and specifications miss important requirements for adapting thelearning process, which was already pointed out for SCORM [Modritscher et al., 2004c] and AICC[Brusilovsky, 2004a]. From the viewpoint of adaptive e-learning, the standardisation process mustcontinue and focus on adaptation issues in order to fulfil these missing requirements.

• Secondly, a few approaches towards e-learning standards tend to accomplish others, as for exampleIMS is supported by a broad range of research institutes and commercial partners [IMS, 2007b] orSCORM has a strong developer community and is applied in many projects and supported by manylearning management systems [Jones, 2002].

• Thirdly, researchers are aware of the fact that content models can be a powerful component ofadaptive e-learning. While [Sosnovsky and Brusilovsky, 2005] focus on topic-based knowledgemodelling, the knowledge space theory by [Falmagne et al., 2003] address competencies and itsrelations to tasks. A skill-based model can be of particular advantage for adaptive assessment of


domain knowledge, calculation of the best-fitting path through the course, providing backgroundinformation and so forth.

• Fourthly, a trend towards adaptation of learner-centred aspects is recognisable in the field of e-learning. For example, [Stash et al., 2006] extended the AHAM reference model as well as theAHA! system with pedagogical aspects like the learning style. Therefore, specifications about thelearners – such as IMS “Learner Information Packaging” (LIP) [IMS, 2007c] or IEEE “Public andPrivate Information” (PAPI) [IEEE, 2001] – as well as connections between learning content andlearner characteristics are targeted by research projects.

As a conclusion it can be stated that many pedagogical and didactical aspects can be described withthe standards and specification drafts available at this time. Nevertheless, there still are a lot of openresearch issues and, additionally, more development work has to be done on e-learning specifications inorder to fulfil requirements of adaptive e-learning. Thus, the following section outlines those requirementsrelevant to adaptation of online learning.

5.2 Requirements for standards to support adaptive e-learning

In accordance with the formal model of adaptive e-learning introduced in the theoretical part of the disser-tation, four main categories for requirements for adaptive e-learning standards can be identified. The firstone comprises all aspects of describing the learning content itself. The second one deals with pedagogicalissues, while the third one addresses didactics. The fourth and final category is about adapting the learningprocess according to the three methods introduced in section 4.4.

5.2.1 Requirements for learning content

The first part of the FORMABLE model dealt with learning content. Generally, learning content comprisesassets – atomic elements like a picture, a paragraph, etc. – and learning objects which define a digitalresource – an asset or an aggregated object – that is used to support learning [DCMI, 2002]. As theFORMABLE model is not aware of aggregating learning objects from assets, atomic parts of the contentare subsumed under the term “knowledge artefacts”.

Yet, the content model of FORMABLE pointed out, that content can be seen as a set of artefacts andis related to a certain context and a certain domain. Therefore, it is necessary to define the relevancy ofartefacts for given situations and concepts, in practice they are linked to learning objectives, which can beunderstood as tuples of a situation and a concept. Moreover, a learning objective can also be dependenton others, as teachers have to define and consider didactical dependencies. As a result, artefacts of thelearning content also underlie dependencies, as is formally described for example by knowledge spacetheory and its application [Albert and Hockemeyer, 1997].

Derived from these theoretical considerations, the following concrete requirements for learning con-tent can be manifested here:

(A1) Defining different types of assets (e.g. text, picture, audio, video, a hyperlink or even a link to aknowledge domain or concept)

(A2) Supporting different types of learning objects (e.g. content, exercises, examination, etc. and anycombination of these types)

(A3) Providing different levels of detail for a learning object (e.g. to address different levels and types oflearning objectives)

5.2. Requirements for standards to support adaptive e-learning 69

(A4) Separating content and presentation for a learning object and offering different visual representationvariants (e.g. for a certain device, browser or bandwidth)

(A5) Creating a learning object (knowledge artefact) through aggregation of different assets

(A6) Modelling knowledge domains and their concepts including overlapping domains or concepts

(A7) Modelling contexts and their situations including overlapping contexts and tasks

(A8) Modelling tuples of concepts and situations and dependencies between them, regarding the preven-tion of a dependency loop

(A9) Mapping a learning object to concepts of domains and contextual situations

Aspects of structuring the course, for example into modules, as well as the visualisation of suchstructures are, from the viewpoint of adaptive e-learning, not of primary interest and, thus, treated as anLMS feature (see also next chapter). On the other side, these requirements are primarily focussed onstandards and specifications describing learning resources, such as IEEE LOM, Dublin Core Metadata[DCMI, 2007], IMS Content Packaging [IMS, 2007a], etc. and content models, like IMS CompetencyDefinition [IMS, 2007f] or specifications from the field of concept modelling and topic maps.

5.2.2 Pedagogical requirements

As the second part of FORMABLE comprises the pedagogical model, standards have to consider suchaspects. This model deals mainly with learner characteristics, but the environment could also be of inter-est. Consisting of a mapping from learners, concepts and situations to states, the pedagogical model looksrather simple, but allows the definition of any possible combination of these three dimensions. Further-more, this approach also points out that assessment of such pedagogical states is the primary target in thefield of pedagogy, which of course is not always possible or easy, as can be manifested for online learning.

From the more practical viewpoint, the following requirements for e-learning specifications can bederived from FORMABLE’s pedagogical component:

(B1) Defining static and dynamic information attributes of a learner

(B2) Providing management (like storage, deletion or update) of attributes in real-time, for example theactual constitution

(B3) Supporting an enhanced learner tracking and modelling (e.g. observing the learning process, thepaths through the courses, all learning objects and assets viewed or the learner’s constitution)

(B4) Mapping a learning object to a learner’s characteristics (e.g. language, accessibility, learning style,multiple intelligence, etc.)

Overall, pedagogical requirements for standards mainly deal with user profiling specifications, suchas IMS LIP or IEEE PAPI. Nevertheless, the second and third requirement also is a recommendation forXML-based formats due to the fact that they support the provision of a data object model for real-timeoperations. Further, environmental states are also often included within such user profiling specifications.Concluding this subsection, it has to be outlined that pedagogical aspects described in the user profile areoften utilised as adaptation information, for example by means of a learner model, in the field of adaptivee-learning.


5.2.3 Didactical requirements

As LMS-centred aspects of FORMABLE’s didactical model are examined in the next chapter, the fol-lowing requirements for standards are focussed on here: On the basis of the content and the pedagogicalmodel, standards need to deal with describing objectives, learning activities and instructional sequences.Furthermore, FORMABLE also comprises the idea of determining the relevance of instructions for givenlearning objectives, defining the suitability of instructions for learners and assessing the learning progressaccording to given mastery levels.

Therefore, from a didactical viewpoint a specification has to fulfil the following requirements:

(C1) Allowing to change the order of the instructional sequence (of all instructions not visited yet)

(C2) Providing different types to sequence instructions (e.g. linear, conditional branches, loops, etc.)

(C3) Allowing the insertion of instructions into the instructional sequence (again under the restrictionthat this new instruction cannot be inserted right after one already visited by the learner)

(C4) Defining pre and post-conditions for instructions (e.g. according to learner characteristics, certainsituations and concepts or a given learning objective)

(C5) Assessing mastery level of learners applying adequate activities (e.g. quizzes, submission tasks,etc.)

(C6) Mapping instructional sequences to learning objectives (e.g. didactical strategies along a learningparadigm)

(C7) Mapping instructional sequences to pedagogical states (e.g. learning units suitable for differentlearning styles)

As can be concluded so far, the FORMABLE model is useful to describe the first five requirementson the basis of the instance variables and the last two ones on level of whole courses. In practice, thedidactical requirements concern specifications like IMS Learning Design. Yet, there are also requirementswhich build up on the last three subsections and focus on adaptation of the courseware.

5.2.4 Requirements concerning the adaptation process

This category of requirements for adaptive e-learning standards is about the three methods to adapt theonline learning process outlined in section 4.4: (1) adaptation of instructions, (2) adaptation of the instruc-tional sequence and (3) adaptation through providing additional instructions. Therefore, requirements forstandards mainly refer to the requirements of the last subsections and can be summarised as follows:

(D1) Defining rules observing pedagogical and didactical states and models (A6-A9, B1-B4, C4-C5) andtriggering adaptation of instructions (A1-A5)

(D2) Defining rules observing pedagogical and didactical states and models (A6-A9, B1-B4, C4-C5) andtriggering adaptation of the instructional sequence (C1-C2)

(D3) Defining rules observing pedagogical and didactical states and models (A6-A9, B1-B4, C4-C5) andtriggering the insertion of new instructions (C3)

These three types of rules already point out which requirements concern the adaptation information(mappings, models and state assessments) and which requirements deal with the so-called adaptors. Ex-amples of specifications in this scope resulted by various projects and researchers, for instance in theMUSE project [Carmagnola et al., 2005], the KOD project [Sampson et al., 2002b], by IMS LearningDesign [IMS, 2007d] or the LMML specifications [Suß and Freitag, 2002].

5.3. Inspection of current specifications 71

5.3 Inspection of current specifications

On the basis of the requirements manifested in the last section, existing standards and specifications areexamined on fulfilling or missing these requirements. Therefore, selected examples of standards andspecifications are inspected for four commonly-known areas. The first one is about describing learningcontent on the level of assets. The second one deals with user profiling specifications in the field of e-learning. The third one comprises specifications describing didactical aspects. Finally, the fourth one aimsat specifying adaptation rules.

5.3.1 Content packaging

Two prominent examples of describing assets and learning objects are the Dublin Core metadata elementset as well as the IEEE Learning Objects Metadata standards. As metadata standards for digital resourceswere targeted years before e-learning specifications arose, both DC and LOM are already approved as stan-dards (see [NISO, 2001] and [IEEE, 2002]). Therefore, these two standards are, after a short introduction,evaluated on the basis of the requirements for learning content.

The Dublin Core metadata [DCMI, 2007] set can be seen as a standard for describing cross-domaininformation resources, i.e. by means of a convention for metadata for digital objects. The standard consistsof elements on two levels: On the one side, the simple element set comprises attributes like title, creator,subject, description, publisher, etc. On the other side, the qualified DC element set stands for an ongoingdevelopment process to extend or refine element set.

Concerning the requirements for learning content from the last section, DC metadata would allowthe definition of different types of assets (A1) and learning objects (A2), using the type-field and customvalues. Further, determining between different levels of detail (A3) would also be realisable, extendingthe element set or misusing a simple field, while the separation between content and presentation (A4)could only be implemented using a relation between two objects described by DC. Contrary to this, thismechanism (defining relations between objects) would enable the aggregation of learning objects (A5),any kind of model, for example a domain, a context or a skill model (A6-A8) and even the mappingbetween skills and learning objects (A9). Yet, describing skills would require other specifications, likeIMS Competency Definition [IMS, 2007f].

The Learning Objects Metadata comprises XML or RDF-based data model for the description oflearning objects. The standard itself arose from the Learning Resource Meta-data specification of the IMSGlobal Learning Consortium (IMS LRM) and was influenced by the ARIADNE research project [Duvalet al., 2001]. [IEEE, 2002] summarises the nine categories of the LOM standard, namely the General,Lifecycle, Meta-Metadata, Technical, Educational, Rights, Relations, Annotations and Classification cat-egory. Nowadays, LOM is the state-of-the-art standard to be considered by learning objects repositories,as shown for example by [EducaNext, 2007], [Singer, 2005] or [C2k, 2003]. Further, LOM is also part ofcommonly-known sets of e-learning specifications, like SCORM.

As the LOM standard is very similar to Dublin Core Metadata – there is also a mapping of elementsbetween these two standards – the requirements A1 to A9 are fulfilled according to the DC standard.Moreover, LOM elements are structured much better, with the nine categories mentioned above and itdoes not have to be extended to fulfil a requirement (A3). Although LOM allows the mapping betweenresource and skills, it misses, adequately for DC, a way to describe competencies. As a conclusion, LOMprovides nearly all necessary elements to allow adaptation of the level of instructions. For missing aspects,other specifications like IMS Competency Definition can be applied.


5.3.2 Learner profiling

With respect to the pedagogical requirements (B1-B4), this subsection addresses standards and specifi-cations in the field of user profiling. For example, “Public and Private Information” (PAPI) and IMS“Learner Information Package” (LIP) are inspected in order to fulfil pedagogical requirements.

The “Public and Private Information for Learners” (PAPI Learner) Standard [IEEE, 2001] comprisesa specification which aims at the data interchange of learner information among learning management sys-tems. One important feature of this specification is the logical division of six types of learner information:(1) personal information, (2) preferences, (3) security, (4) relations, (5) performance and (6) portfolio.

The IMS “Learner Information Package” (LIP) [IMS, 2007c] also deals with an XML-based descrip-tion of learner information. Yet, it consists of much more categories, namely identification, goal, “qcl”(qualifications, certifications and licenses), activity, transcript, interest, competency, affiliation, accessibil-ity, security-key and relationship. Nevertheless, both specifications are more or less equal to each other,as [Chatti et al., 2005] state that the six PAPI sections can be mapped to the IMS LIP categories and viceversa.

Referring to the pedagogical requirements, both specifications allow the definition of static and dy-namic attributes of learners (B1). Further, the management of attributes in real-time (B2) is also supporteddue to the fact that PAPI and LIP are based on XML and the generation of a data object model can beeasily achieved. Finally, exact learner tracking (B3) and mappings to learning objectives or various char-acteristics (B4) are supported by different categories, such as PAPI’s performance and portfolio section.As a conclusion, pedagogical requirements can be considered to be fulfilled, although LIP and PAPI areprimarily standalone specifications and, therefore, not integrated in commonly-known specifications suchas SCORM.

5.3.3 Courseware

As the didactical requirements for adaptive e-learning standards address instructional design, standardsand specifications like AICC “Web Based Computer-Managed-Instruction”, IMS “Simple Sequencing”and “Learning Design” and SCORM as a whole are of interest for this subsection.

AICC’s Guidelines and Recommendations [AICC, 1998] include two sections, namely the AGR006(CMI) and AGR010 (web-based CMI), which focus on web-based and computer managed instructions.[Bergstrom, 2001, p. 133ff] introduced the concepts of describing a course hierarchy and the instructionalsequence for the first time. Further, this guideline also deals with (learning) objectives and prerequisitesfor instructional units. On the other side, the guideline also includes aspects of content packing, as men-tioned in the subsection before the last one. Nevertheless, these two AGRs had a strong impact on otherspecifications, for example IMS Simple Sequencing.

IMS Simple Sequencing specification defines “a method for representing the intended behavior of anauthored learning experience such that any learning technology system (LTS) can sequence discrete learn-ing activities in a consistent way” [IMS, 2007g]. Thus, it specifies the possible paths through a course bymeans of didactical dependencies and all possible sequences of instructions. IMS Simple Sequencing alsoallows the definition of so-called objectives which describe a learner’s state towards a learning objective.Objectives can be given from outside, for example a learner profile, or be assessed by certain instructions.

Therefore, instructions for assessment have also to be specified, which can be realised with a speci-fication like IMS “Question & Test Interoperability” [IMS, 2007e]. Additionally, IMS Learning Design[IMS, 2007d] aims at describing instructional design in connection with pedagogical issues. The LearningDesign specification can be understood as generic and flexible language to support different pedagogicalparadigms, for example didactical strategies based on the learning theories relevant for e-learning (seesection 3.1). This extension was necessary due to the fact that IMS specifications lack possibilities to

5.4. A standard-based approach to adaptive e-learning 73

determine pedagogical issues for courseware.

Finally, the Sharable Content Object Reference Model (SCORM) by Advanced Distributed Learning[ADL, 2007a] comprises a set of standards and specifications build up to create standardised coursewarein practice. Therefore, SCORM utilises standards like IMS Content Packaging and IEEE LOM for contentpackaging as well as IMS Simple Sequences to determine learning paths through the course. Yet, it stillmisses the possibilities of learner profiles and learning design.

As a conclusion, SCORM in combination with IMS LIP and IMS Learning Design for example wouldrequire most of the didactical requirements manifested in the last section. While instructional sequenc-ing would allow the definition of different paths through a course (C1, C2) and specifying pre and post-conditions for each instruction on the basis of achievement levels given from user profile or via assessment(C4, C5), Learning Design would provide the mapping of instructional sequences to pedagogical states(C7). Nevertheless, retrieving and inserting new instructions (C3) is not directly supported at all by ex-isting specifications and standards and would require workarounds to be realised, as shown in the nextsection.

5.3.4 Adaptation rules

Adaptation rules in general have not been targeted intentionally by standardisation efforts so far. Thus,specifications considering such aspects arose rather by chance, as for instance IMS Learning Design aimsat adaptation aspects, although adaptivity or adaptability of courseware is not a primary goal of this spec-ification and, further, not even mentioned in this context. Yet, various research projects in the area ofpersonalised and adaptive e-learning build up such specifications, partially on the basis of existing oneslike in the KOD project [Sampson et al., 2002b], partially new ones on the basis of OWL (MUSE project[Carmagnola et al., 2005]) or XML (LMML [Suß and Freitag, 2002]).

Beside these proprietary specifications resulted from research projects and concerning the require-ments for adaptation rules (D1-D3), only the IMS Learning Design specification is aware of the definitionof rules to trigger adaptation of instructions or the instructional sequence. Inserting new instructions intoan existing course, which is targeted by retrieval-based techniques, has not been an issue in the stan-dardisation process so far and has to be implemented with workarounds, as pointed out in the approachintroduced in the following section.

5.4 A standard-based approach to adaptive e-learning

Due to the necessity to standardise courseware as essential criteria for transferability and reusability ofcontent [Qu and Nejdl, 2002], e-learning standards and specifications are also of relevance for the AdeLEresearch project. Therefore, this section briefly introduces the AdeLE research project, describes thedecision for a certain set of specifications more closely and outlines enhancements of these specificationsto fulfil the requirements given by the project.

5.4.1 The AdeLE project and its requirements for learning content

The term AdeLE [AdeLE, 2006] which is the abbreviation for “Adaptive e-Learning with Eye-Tracking”stands for a research project which ran from 2003 to 2007. From the theoretical viewpoint, this projectaims at the realisation of an adaptive e-learning environment, the AdeLE prototype, considering theoreticalaspects given by the FORMABLE model. From the practical viewpoint, AdeLE comprises a technology-based solution exploiting novel methods for retrieval-based instructions and fine-grained user profilingbased on real-time eye-tracking and content-tracking information.


Derived from section 5.2, the requirements for standards towards adaptive e-learning have to be sup-ported by the specifications utilised for the AdeLE prototype. Nevertheless, the application of eye-trackingtechniques brings up special needs. Firstly, it is necessary that not only whole instructions, but also in-structional parts such as images or text passages are connected with didactical objectives. Therefore,such passages might be annotated by the teacher with objectives like “to read” or “to learn”. Secondly,the idea of applying retrieval-based techniques also requires workarounds and enhancements of existingspecifications, as mentioned before.

As a conclusion from the inspection of current e-learning specifications outlined in the last section,the application of the specifications and standards of ADL’s Sharable Content Object Reference Model(SCORM) can be considered as a good basis for describing adaptable courseware. Also highlighted in thelast section, these specifications do not fulfil all requirements for adaptive e-learning. As a result, theseexisting specifications have to be extended on the level of instructional sequencing and the instructionallevel. Further, a few workarounds are necessary in order to utilise SCORM for the AdeLE prototype.

5.4.2 Enhancements in organisation and content

The overview of SCORM given in figure 5.2 visualises the main elements of the SCORM specifications:Firstly, mark [a] labels the “Organization”, which describes the structure of a course with so-called itemsand allows the determination of different instructional units like modules, instructions, etc. Secondly,mark [b] comprises the reference from each item to exactly one instruction, namely a sharable contentobject (SCO) or, in former chapters, also called a “knowledge artefact” or “learning object”. Thirdly,mark [c] is about aggregating assets to SCOs. And finally, mark [d] deals with instructional sequencing,which is realised by IMS Simple Sequencing within SCORM.

Figure 5.2: Enhancing SCORM’s structuring and content packaging specification

From the viewpoint of adaptive e-learning, the “Organization” [a] is more or less irrelevant, while thereference between items and SCOs [b] could be loosened to realise retrieval based instructions. Thus, morethan one SCO could be addressed or some query term could be defined in order to retrieve instructionalcontent from the course package or even another learning object repository. Further, the aggregation ofSCOs from assets [c] could be, on this level, enhanced to retrieve the SCOs and assets of a learning objector to generate the learning object from other SCOs or assets. Yet, this approach would require either aretrieval component or an aggregation engine for learning objects.

5.4. A standard-based approach to adaptive e-learning 75

Instructional sequencing [d] allows the adaptation of a given path through the course, for example onthe basis of objectives or pedagogical states. In any of these cases, connections to a learner profile orpre-defined competencies could allow an adaptive component to adapt the courseware on the basis of thestandards and specifications of the commonly-known SCORM. Overall, the suggestions mentioned herewould enhance the adaptability of courseware towards some requirements for learning content (A5-A9)and all pedagogical (B1-B4) and didactical requirements (C1-C7), all introduced in section 5.2.

5.4.3 Enhancements in assets

Describing assets in SCORM is based on the IEEE LOM specification. Therefore, it is possible to use allattributes of the nine categories of LOM. The educational elements are of particular interest for adaptationaspects, as they include didactical and also pedagogical issues like the resource type, interactivity level,difficulty level, the context and so forth. Further, it is also possible to describe relations between assets,such as “is part of”, “is related to”, etc. Therefore, it is possible to define the format of an asset, as shownby mark [a] in figure 5.3. On the other side, assets can also be modelled as a set of documents. In the caseof mark [b] it is also thinkable that the visual representation of an instruction can be adapted by combininga XML-resource with different XSLT-files.

Figure 5.3: Enhancing SCORM’s asset specification

To fulfil the other requirements for learning content (A1-A4), an extension of assets is not necessaryat all. Yet, it would be required to utilise the possibilities of LOM, particularly for describing relationsbetween resources and exploit these content model. As LOM mainly focuses on didactical aspects, ped-agogical issues can be realised with other mechanisms, like applying IMS Learning Design. Generally,these two layers of the SCORM specifications already fulfil important requirements for standards to sup-port adaptive e-learning.

5.4.4 Further suggestions for the AdeLE approach

As SCORM does not support user profiling so far, it is necessary to adopt an existing XML-based spec-ification like IEEE PAPI or IMS LIP. According to the pedagogical requirements, the chosen specifica-tions have to consider domain-specific information like domain knowledge, records of learning behaviour,records of evaluation and assessment [Brusilovsky, 1994] plus domain-independent information like cog-nitive attitudes, motivational states, background knowledge, multiple intelligences, learning styles, etc.[Lane, 2000]. Furthermore, it is important that this specification is performs well to allow real-time usertracking.


Regarding requirements for learning content, a standard needs to support modelling of knowledgedomains and contexts for e-learning. Therefore, another specification like the IMS Competency Definition[IMS, 2007f] has to be adopted. Furthermore, the specifications for learning content and sequencing oughtto be connected to the learner’s characteristics, knowledge domains and contexts. Again, SCORM doesnot include these aspects yet, so a specification like IMS Learning Design is required.

One very special requirement of the AdeLE project, which is derived from the application of eye-tracking technology, deals with defining learning objectives for elements within an instruction. This re-quirement could have been realised by describing assets and aggregating them to instructions. As theAdeLE prototype does not include an aggregation engine for instructions, an own tool was developed inorder to define inner-instructional objectives.

Figure 5.4: “Semantic TAGging Editor” for inner-instructional objectives

The so-called “Semantic TAGging Editor” (STAGE) which was mentioned firstly in [Garcia-Barrios,2006] and is implemented as a Firefox plugin allows the selection of HTML-elements and the annotationof them with states like “to read” or “to learn”. Figure 5.4 displays the STAGE tool in front of an openedinstruction, which was partially annotated with elements to be learned (red colour) and elements to be read

5.5. Conclusions 77

(yellow colour) by the learner. Contrary to all other adaptation information, these objectives are not storedas metadata, but within the instruction itself. How it is utilised is shown in chapter 7, which describes theAdeLE prototype in detail.

Concluding this section, two further issues have to be outlined here: On the one side, the SCORMmanifest file is based on XML, which enables that the extensions suggested in this section can be easilyrealised by modifying the mark-up definition, for example by adding entries of the IMS Learning Designor IMS LIP into SCORM’s manifest file. Further, XML is also advantageous for building up a data objectmodel (DOM) for real-time data processing. On the other side, the standard-based approach introducedhere also supports retrieval-based instructional creation, as shown with the idea of a Dynamic BackgroundLibrary in chapters 7 and 9.

5.5 Conclusions

Against the background that standardisation is a contradiction to personalisation this chapter pointed outthe necessity of utilising standards and specifications and, therefore, suggests a standard-based approachtowards adaptable courseware. After giving a short overview of the standardisation process in the field ofe-learning and outlining a few speculations on further development of such standards, this chapter provedthe usefulness of the FORMABLE model of the dissertation’s theoretical part for the first time. In thecontext of the rather practical approach to define requirements for e-learning specifications to supportlearner-centred adaptivity, FORMABLE was applied to derive these requirements.

As a result, four categories of such requirements were identified. The first one focuses on the learningcontent, suggesting the description of instructional content, the knowledge domain and learning contexts.Secondly and thirdly, pedagogical and didactical issues have to be described in order to adapt the learningprocess to these aspects. Fourthly, requirements concerning adaptation rules deal with adaptation methods,such as adaptation of instructions, the instructional sequence or inserting new instructions. All in all,specifications have to fulfil these requirements in order to support adaptive e-learning.

An inspection of current e-learning standards outlined that commonly-known sets of specifications,like SCORM, do not fully support aspects of adapting the online learning process. Thus, the AdeLEproject team attempted an approach to utilise the specifications and standards of SCORM and consid-ered missing functions with other specifications, for example IMS LIP or IMS Learning Design, andworkarounds, so that the final prototype implements as many features of the FORMABLE model as pos-sible. In addition to adaptable courseware, the learning management system also plays an important roleif the learning process is to be adapted. The next chapter therefore deals with an ideal adaptive e-learningenvironment.

Chapter 6

An Ideal Environment for Adaptive E-Learning

“ Idealism increases in direct proportion to one’s distance from the problem. ”

[ John Galsworthy ]

To describe an ideal adaptive e-learning environment, this chapter is structured as follows: First of all,section 6.1 summarises commonly-known methods and techniques to adapt the online learning process.Then, an ideal model of an adaptive e-learning environment is depicted on the basis of functional require-ments (section 6.2) and an architectural design (section 6.3). In these three sections, the FORMABLEmodel is utilised to deduce various facts. Concluding this chapter, existing projects and solutions areinspected in section 6.4 with respect to the ideal adaptive e-learning environment.

6.1 Methods and techniques for adapting the learning process

According to [DeKoch, 2000, p. 19], an adaptation method is determined “by an adaptation idea definedat conceptual level”, while a technique is defined “by a user model representation and an adaptation algo-rithm”. Methods and techniques for adaptation in e-learning environments mainly deal with the elementsof the online learning process which can be adapted. Referring to FORMABLE, adaptation methodscan be divided into three categories: (1) the ones adapting the instructions itself, (2) the ones adaptinginstructional sequences and (3) the ones adapting an online course by inserting new instructions.

6.1.1 Adaptation of instructions

This category of methods and techniques comprise the idea that an instruction itself or parts of it canbe adapted in some way. Therefore, it has to be distinguished between fragment and page variants. Onthe other side, researchers differentiate between content adaptation and adaptive presentation. These twodimensions serve as a basis for giving an overview of methods and techniques in the scope of adaptivee-learning as well as for categorising them into this or into one of the next two subsections.

With respect to [Brusilovsky, 1996], content adaptation methods on an instructional level compriseadditional, prerequisite and comparative explanations, explanation variants as well as sorting, hiding andannotating instructional fragments. Yet, it has to be stated that additional explanations have to be treatedonly on a fragmental level to fit into this category. Further, adaptive presentation with methods like multi-languages or layout variants can also be categories here. These methods can be of benefit for the learningprocess if applied on the basis of didactical or pedagogical principles.

79

80 6. An Ideal Environment for Adaptive E-Learning

Addressing techniques on this adaptation level, two types can be outlined here: On the one side,adaptive presentation techniques include the provision of instructions on different levels of detail, in amultimodal way, following different layouts or adapting navigational elements of the user interface toimprove orientation. Concerning layout variants, the style-guiding technique is often applied. On theother side, instruction level adaptation also utilises techniques like conditional text, stretch text, page orfragment variants and frame-based approaches.

[Henze, 2000, p. 15] outlines that conditional text allows the definition of inner-instructional depen-dencies between fragments; stretch text aims at describing keywords more comprehensively; and frame-based techniques (on fragmental level) determine certain rules to present fragments according to a specialorder, for example given for didactical reasons. Further, one technique of adaptive navigation, namely linkannotation, would also match this category, because the instruction or at least some navigational elementshave to be adapted.

With respect to FORMABLE, the methods and techniques of this category can be subsumed by thetwo operations “AdaptInstructionByDidactics” and “AdaptInstructionByPedagogy”. Therefore, the exacttechnique applied on an instructional level can be described by one of these methods, as they are restrictedto selection and presentation of learning content. Furthermore, the formal approach to the last chapterdefines an instruction as a pair consisting of a knowledge artefact (the content) and a so-called activity,which represents certain features of the learning management system.

Examples of such activities are, besides passively presenting content, assessment methods like quizzesor tasks, feedback mechanism like form-based questionnaires and even collaborative tasks, such as dis-cussions or chats. Therefore, FORMABLE not only deals with adapting the content itself, but is alsocapable of varying learning activities. Additionally, it is more relevant to examine dependencies betweeninstructional design (including for example some adaptation technique) and the didactical and pedagogicalmodels behind the system and, particularly, the effect of adaptation of the learning process.

6.1.2 Adaptation of the instructional sequence

The instructional sequence can be defined as the learner’s path through a course. In the FORMABLEmodel, this order is given by the course itself, which is specified as a sequence of instructions, but canbe adapted by the methods “AdaptSequenceByDidactics” and “AdaptSequenceByPedagogy”. Referringto the learning paradigms in section 3.1, the range of instructional sequencing can reach from a strictorder keeping the learner on exactly one pre-defined path to a course in which the learner can select eachinstruction directly.

Restrictions of the instructional sequence are mainly given by didactical aspects, for instance themacro-adaptive instructional approach in section 4.1 outlines the importance of dependencies between in-structions or pre-assessments for instructional units. Theoretical approaches for adaptation of instructionalsequences were for example introduced in section 4.3, whereby the knowledge space theory by [Albertand Hockemeyer, 2002] is an example of the definition of didactical dependencies. On the other side,[Karampiperis and Sampson, 2005] examine all possible combinations of sequences according to a mediaspace and a domain model as well as learner observations.

[DeKoch, 2000, p. 18] considers the instructional sequence (the structure) as “the organisation ofthe content specification as to which content items will be visited and how they will be visited throughnavigation”. Therefore, some methods and techniques of adaptive navigation can also be assigned to thiscategory of adaptation. [Brusilovsky, 1996] enumerates adaptation methods like global guidance (shortestpath) and local guidance (best fitting link) as well as orientation support addressing the user’s knowledge,goals and a global view of the system.

On the other side, [Conlan, 2005, p. 35ff] summarises techniques such as relevance (e.g. given bydidactical aspects), direct guidance (next-button with best fitting instruction), link ordering (link-list sorted

6.2. Functional requirements 81

by relevance) and link hiding (removing irrelevant links). The frame-based technique by [Henze, 2000,p. 15] would also fit into this category, if treated at page level. Generally, these techniques can be realisedby re-ordering the instructional sequence, which has a further effect on the user interface, for example onnavigational elements.

Finally, it has to be mentioned that adapting the sequence of instructions is only possible for instruc-tions not visited yet. Past instructions should not be re-ordered at all, so that the adaptation process fulfilsthe requirement of scrutability and the learning history is reproducible for both the learner and the instruc-tor. This fact is considered by FORMABLE by the pre-condition in the two methods referred to in thissubsection.

6.1.3 Adaptation through insertion of new instructions

The last category of adaptation methods and techniques deals with the idea that for different reasons newinstructions can be inserted in the course sequence. Possible reasons can be derived from didactics, forexample a learner requires another explanation, or from pedagogy, for instance instructional alternativesare added in order to meet the principle of the dual coding theory.

The FORMABLE model fulfils this requirement with the two methods “InsertInstructionByDidactics”and “InsertInstructionByPedagogy”. As these methods could also deal with repeating instructions, pre-conditions about forbidding the insertion of past instructions is allowed. Contrary to re-ordering theinstructional sequence, this instruction is duplicated and not moved.

Methodologically only a few concepts can be identified for this category, all given for didactical rea-sons. The method of additional explanation at page level was already excluded in the former subsection,but fits for this one. Another approach comprises didactical strategies for repeating instructions, for exam-ple because certain activities have to be experienced multiple times, or assessment identified knowledgegaps.

From the viewpoint of techniques, the insertion of instructions can reach from an instructor’s providingstatic links to the idea of retrieving all instructions from one or more learning object repositories, as seenfor example by the Knowledge Sea portal in [Brusilovsky, 2004b]. Another interesting approach in thiscontext is the idea of the Dynamic Background Library, described in detail in [Dietinger et al., 1999] and[Modritscher et al., 2005] and evaluated in chapter 9 of this work. The technique used in this approachcan be positioned in the middle of static link-lists and courses consisting fully of retrieved instructionalcontent.

As a conclusion to this section, it can be stated that from the viewpoint of research on adaptive e-learning it is obviously much more relevant how the system behaves, on the basis of which informationadaptation takes place and what the didactical result of the adaptation is. While adaptation methods areslightly related to the theoretical models behind an adaptive e-learning environment, techniques are onlyinteresting, if they are evaluated for a certain didactical or pedagogical model, which often is not the caseas stated for example by [Park and Lee, 2004]. Thus, the FORMABLE model can be understood as atheoretical framework for examining exactly these dependencies between techniques and the didacticalmodel within an adaptive e-learning environment.

6.2 Functional requirements

While aspects of adaptable courseware were already examined in the last chapter, this section deals withfunctional requirements of e-learning systems in order to adapt online courses.


6.2.1 State-of-the-art e-learning features

From a traditional viewpoint of e-learning a lot of research work has been done already and, further,various existing platforms document which features and user requirements are necessary to realise onlinecourses. For instance, [Dietinger, 2003, p. 41ff] groups functional requirements for e-learning systems intothree major groups: (1) learning management systems (LMS), (2) learning content management systems(LCMS), (3) learning and tutoring support management (LTSM).

Learning content management comprises all functions to create and maintain courseware. Thus, theserequirements mainly deal with authoring tools, learning object repositories, e-learning standards and col-laborative features to create content. As these aspects are not of relevance to adapting the learning processand e-learning specifications and standards encapsulate the authoring process of learning content, learningcontent management will not be treated closer here.

On the other side, LMS and LTSM primarily focus on the learning process, whereby learning man-agement sums up all functions concerning the learner interacting with the course material and learningand tutoring support is about communication and collaboration amongst learners and with the teacher.Features of an ideal LMS already include particularly important aspects of adapting the learning process:

(A1) The learner portal stands for the user interface of an e-learning platform. Adaptability of the userinterface, i.e. the ability to customise the portal, can be considered to be state-of-the-art.

(A2) A registration module allows learners to access courses. Further, it handles billing issues and pro-vides notifications, policies and information on the competencies to be achieved by the courses.

(A3) The learner profile component is aware of features to manage information about the learner. Inaccordance with the pedagogical requirements of section 5.2, a learner profile comprises static in-formation (the learner records), tracking the learning history, personal performance and skill gapanalysis and different reporting functions for teachers.

(A4) The course presentation engine delivers the instructions to the learner and, ideally, provides differentnavigational elements for information visualisation purposes. The most commonly-known elementsare, beside suspending or exiting a course, a previous and a next button as well as a tree-view forvisualising the course hierarchy.

(A5) An assessment component allows the assessment of the learning process, for example by instruc-tions like quizzes or tasks, form-based questionnaires or other activities.

(A6) Administrative features deal with typical tasks for course administrators or teachers, for instancethe course management, system configuration or determination of different roles.

Additional to LMS features, it is recommended to implement the following features of learning andtutoring support management:

(A7) Communication tools allow learners to communicate with each other or with the teacher. Examplesof asynchronous features are emails, discussion groups, etc., while synchronous communicationcould be realised with instant messaging or video conferencing [Rollett, 2003, p. 105ff].

(A8) The collaborative group of requirements sums up all learning activities which are interesting forgroup work. For example, groupware tools like task lists, a calendar, shared workspaces, etc. arementioned here. Younger streams also comprise social software [Farzan and Brusilovsky, 2006] orconcepts of Web 2.0 [Rollet et al., 2007].

6.2. Functional requirements 83

Referring to the theoretical part of this work, these two requirements aim at the constructivistic-collaborative approach to section 4.1, while the other functions are mainly related to instructional designoutlined in section 3.3.

6.2.2 Functional enhancements for adaptive e-learning

Examining e-learning platforms under the view of adaptive e-learning, the FORMABLE model can beutilised again in order to extend the eight requirements mentioned above. Aspects of a content model canbe identified in (A2), as the registration module has to include information about competencies achievableby the courses. Such information could be of benefit for adaptive assessment (A5), particularly if a fine-grained skill model retrieved from the learning content is applied [Albert and Hockemeyer, 2002].

Pedagogical issues are mainly comprised by the learner profile component (A3), which should bebased on a commonly-known standard as recommended in the last chapter. Additionally, the assessmentcomponent (A5) should support not only the assessment of knowledge-dependent aspects, but also ofpedagogical states. For instance, a form-based psychological test like the so-called VICS v2.2b program, apsychological test to retrieve the WAVI-factors [Peterson et al., 2003], might be of interest, if the cognitivestyle of a learner has to be modelled.

Beside the administrative features (A6), which are not of relevance for adaptive e-learning, all otherrequirement groups concern the didactical aspects. Because the FORMABLE model defines an instructionas a pair of an LMS-feature and a knowledge artefact, the learner portal (A1) and its adaptability can beseen as a typical LMS-feature. On the other side, communication tools (A7) and collaborative activities(A8) have to also be classified as LMS-features which can be combined with instructional content, forexample in the form of an email discussion or a group task on a course’s topic.

Adapting the learning process primarily comprise the course presentation engine (A4) and the as-sessment component (A5), which are responsible for observing the learning behaviour and adapting in-structional delivery according to one or more of the three adaptation categories: adapting an instruction,adapting the instructional sequence and inserting an additional instruction. Therefore, these two groups ofrequirements have to implement the possibilities pointed out with the requirements for adaptable course-ware, particularly the ones concerning didactics and the adaptation process.

The methods and techniques for adaptive e-learning summarised in the last section, are partially con-sidered by the requirements for adaptable courseware. An ideal adaptive e-learning environment has tofulfil these additional functions:

(B1) For adaptation of the instructional level it is necessary to provide an aggregation engine in orderto generate instructions from a set of assets and a description of relations between them on-the-fly. Furthermore, the adaptation of navigational elements and presentation of instructions should besupported by the LMS.

(B2) Adaptive sequencing requires that the data object model of the instructional sequence can be ma-nipulated and personalised for each learner at runtime.

(B3) Retrieval-based generation of instructions brings up the requirement that the LMS has access toinformation retrieval systems. In the context of e-learning, such a system might be a learning objectrepository.

In practice, requirements (B1) and (B2) are already utilised in many products and by many researchers.Approaches to adapt instructions and the instructional sequence is well founded by psychologists underthe term “instructional design” [Spector and Ohrazda, 2004] and seem to be widely spread as shown infield-specific conferences [Wade et al., 2006], [Bauer et al., 2005], etc. Retrieval-based instruction (B3) is


examined and applied as well, mostly under the term constructivistic or exploratory learning or in the fieldof adaptive hypermedia [Brusilovsky 2001]. Exemplary solutions are the Knowledge Sea [Brusilovsky2004] or the idea of the Dynamic Background Library [Dietinger et al., 1999] which is described closelyin chapter 9.

6.2.3 Non-functional requirements

Beside these functional requirements for adaptive e-learning environments, there are also other featureswhich do not directly deal with adapting the learning process, but can be derived from literature or, forinstance, from the field of adaptation systems. In the following, these requirements are briefly outlined:

(C1) From the viewpoint of adaptation systems, it is important that the adaptive behaviour can be under-stood by the user. Therefore, concepts towards scrutability have to be implemented. For example,an adaptive e-learning system could visualise or point out adaptation decisions to the user.

(C2) Another systemic attribute concerns controllability. As adaptivity might be provided by an ownspecialised component, like a multi-purpose adaptive system indicated in section 2.4, a learningmanagement system has to provide an interface, which can be used by such an adaptive componentto adapt the learning process automatically. This interface comprises not only the possibility toexchange data, but also some kind of API to control the behaviour of the LMS.

(C3) In order to evaluate the effects of adaptation, a logging mechanism and intelligent analysis methodsare required. Adaptation decisions and the actual state of the adaptation information have to be writ-ten to log-files, so that researchers can evaluate the usefulness of the system’s adaptive behaviour.

In literature, some other requirements also relevant to adaptive e-learning systems can be observed. Onthe one side, [Dietinger, 2003, p. 47ff] outlines various technical requirements like performance, security,support of standards, usability, etc. On the other side, [Conlan, 2005, p. 97ff], amongst others, addressesthe idea of a distributed e-learning environment based on a service-based architecture. Such an approachstrengthens the idea that e-learning can be understood as tool repository for technology-based learningand teaching [Modritscher et al., 2006a]. Thus, the following section presents an exemplary architecturefor an adaptive e-learning environment.

6.3 Architectural design

Tying up to the trend of encapsulating functions into components leads to an architecture for an adaptivee-learning environment as shown in figure 6.1. This approach aims at developing such technologicalentities providing a certain and well-defined functionality. Moreover, this architecture would also allowthe distribution of these components on different computer systems, for example for performance reasons.In this concrete case a separation of a unique server-sided system and a client-sided part, which can existonce per learner, is already indicated.

6.3. Architectural design 85

Figure 6.1: Architectural design of an adaptive e-learning environment

The most outstanding advantage of this approach deals with the reuse of technological solutions. Forinstance, if a learning management system is given or already in use, it can be used as a starting point andcould be combined with a multi-purpose adaptive component outlined in section 2.4. Then, modellingsystems can be added and, finally, the loop beginning with the presentation of the course material over theuser interaction with the system, the tracking and modelling components to the adaptive engine and thelearning platform can be closed.

As this architectural design gives a first overview of the logical units of an ideal adaptive e-learningenvironment, the following subsections explain the functions of each component in detail and outlineaspects of adaptation systems and e-learning.

6.3.1 E-learning platform and content modelling

The e-learning platform comprises a unique backend system as well as a client-sided frontend for thelearners. These two components stand for the learning management system and the learning and tutoringsupport management mentioned in the last section. Additionally, the repository containing the courseware,a part of the learning content management, can be assigned this functional area, as one of the main tasksof the e-learning platform is the deliverability of the online courses.

From the didactical viewpoint, the e-learning platform includes all aspects of instructional design andsequencing. Yet, content models can also be derived from the learning content as outlined in the finalchapter or shown by approaches utilising text mining and information extraction techniques [Sosnovskyand Brusilovsky, 2005]. Such models have an impact on learner profiles and the adaptive engine.

Addressing adaptation of the learning process, the e-learning platform is the most important compo-nent regarding the adaptable objects of the adaptation process. As outlined with the requirements for anadaptive e-learning system in the last section, the LMS needs to provide many possible learning activitiesand a highly adaptable user interface. Furthermore, it has to manage and provide the courseware, whichis ideally described with commonly-known standards and specifications and allows the aggregation ofinstructions from assets, to rearrange the instructional sequence or to insert net instructions.


As the e-learning platform is the central interface between the learning content and the learner, allpossible logging information is of interest for evaluation purposes. On the one side, teachers could analysethese log-entries in order to get feedback on a course’s quality. On the other side, user interaction can beexamined and, therefore, the effects of adaptation can be measured. Both evaluation types will be appliedwithin a case study in chapter 10.

6.3.2 Learner tracking and pedagogical modelling

Although the user interaction could be and normally is recorded by the e-learning platform, it is rec-ommended to implement the full learner tracking and pre-processing for the learner modelling on theclient-side. When applying new methods or technologies for learning observation, like eye-tracking de-vices, a huge amount of data has to be managed. In order to not overload a central server containingthe whole or parts of the adaptive e-learning environment it is recommended to process these data onthe client-side, to derive facts about learning behaviour by means of some model and to send this rathercompressed information to the pedagogical modeller.

Pedagogical modelling comprises the management and provision of all models concerning the learneror the learning environment. Models not related to the learner or the learning situation should be part ofthe content model outlined in the last subsection. As there are connections between a content model anda learner model, it might be necessary that the pedagogical modeller can communicate with the contentmodeller. Such scenarios comprise primarily the domain model or even the skill model of an online course,which can be exploited for enhanced tracking of the learning progress. The states of the pedagogicalmodel can be updated by learner tracking mechanism on the client-side or assessment activities within thee-learning platform.

In the context of adaptive e-learning, both the content and the pedagogical models comprise the adap-tation information, i.e. the models to be observed to trigger the adaptation. The difference between thesetwo models is closely related to macro and micro-adaptive instructional systems, as explained in section4.1. While a content model must not change during the learning process, adaptation can be triggered be-fore instruction. On the other side, pedagogical states underlie a continuous flow of the student’s learning.Thus, the learning process can also be prescribed on-task.

6.3.3 Adaptive component and logging

The heart of adapting the learning process is the adaptive engine. As indicated with the directed arrowsin figure 6.1, this component exploits the content model and the pedagogical model directly. Didacticalaspects, which are included within the standardised courseware, also have to be considered – whether aspart of the content model or on request via e-learning platform – because they specify which elementsof the learning content can be adapted in addition to the LMS features. For instance, if a course doesnot allow learners to choose between instructions, it is not possible to adapt the instructional sequence bymeans of the best fitting instruction next.

On the other side, the adaptive behaviour is also restricted by the features of the learning managementsystem. Micro-adaptive instructional systems particularly require not only topical pedagogical models,but also the functions outlined in the last section. Adaptation rules have to be provided by some kind ofrepository within the adaptive engine. One outstanding advantage of standardised courseware deals withthe fact that the rules can be defined with respect to these XML-based descriptions of e-learning content.[Conlan, 2005, p. 67] characterises this idea as “metadata-driven approach to personalised e-learning”.

For research and scrutability purposes it is important to log any kind of information about the adap-tation process. Referring to the architectural design depicted in this section, this kind of logging has tobe conducted on the server and the client-side. On the one hand, a scrutable systemic behaviour would

6.4. Inspecting existing projects and solutions 87

explain to the users, why and on which basis adaptation decisions were made. On the other hand, log-entries could help researchers to analyse the effect of the learning process as well as the usefulness ofadaptation. Important questions comprise aspects of the validity of models, particularly of pedagogicalones and adaptation decisions which were useful or useless.

In the field of adaptive e-learning, adaptation decisions have to be evaluated according to the learningoutcomes, as will be indicated in the case study in chapter 10. Further, it remains to analyse whether theadaptation process is well-defined and deterministic. Non-deterministic adaptation might be recognisableby certain patterns of alternating decision sequences which repeat very often. Yet, ill-defined adaptivebehaviour can be confusing for the user, as stated by [DeBra, 2000]. With respect to chapter 4 of thiswork, the benefit of many approaches in the field of adaptive e-learning, for example of various ATI-basedmethods, is still unproven.

6.4 Inspecting existing projects and solutions

After building up requirements for an ideal adaptive e-learning environment and pointing out very flexiblearchitecture, this section inspects exemplary research projects and solutions in this area. Therefore, therequirements for adaptive e-learning standards and environments will be commented on briefly in the nextsubsections.

6.4.1 Early research activities

Researchers dealt a lot particularly with concepts of adaptive e-learning, as already mentioned in chapter4. In the following selected projects of the early research work are examined here.

First of all, [Brusilovsky et al., 1996] introduced ELM-ART (ELM Adaptive Remove Tutor) as anexample of a portal which provides features of intelligent tutoring systems (ITS) via the web. As a result,an intelligent textbook with integrated problem-solving assets is accessible by learners. The system canbe seen as LMS with restricted functionality (e.g. no collaborative elements), but supporting simpleauthoring features such as the possibility to structure the courseware. At the backend, the Interbook system[Brusilovsky et al., 1998] was applied for creating and delivering these adaptive electronic textbooks.From the technological point of view ELM-ART is realised by various modules within a Common LispHypermedia Server (CL-HTTP). E-learning standards were not considered in this approach yet.

Adaptation within the ELM-ART system is mainly restricted to content-based aspects by means ofassessing the learner’s achievements on pre-defined concepts. Further, the assessment of the learner’scharacteristics is realised by personal preferences and online tests. Adaptation techniques comprise navi-gational elements, link annotation and sorting, instructional dependencies, providing additional informa-tion as well as tutorial support via ITSes. Nevertheless, adaptive behaviour on the basis of pedagogicallearner states is not covered by these early attempts at an adaptive educational hypermedia system. AsELM-ART in combination with Interbook is one of the pioneers in the field of adaptive hypermedia, italready provides intelligent tutoring functionality as well as content-based adaptation.

Secondly, the KBS Hyperbook System [Henze, 2000, p. 31ff], which was developed at the Universityof Hannover, is a framework for modelling, organising and maintaining learning material retrieved frominformation space and to be provided for learning in projects. In general, this system is a learning manage-ment system including some authoring functionality, namely a mechanism for indexing the materials. Theoverall system is realised with web technology by means of the Java Web Server. Aspects of standardisedlearning content are not considered at all.

Concerning adaptive e-learning, KBS Hyperbook supports primarily content-related methods of themacro-adaptive instructional approach, which comprises the assessment of the learner’s domain knowl-


edge and adaptation of course content towards this information. Therefore, all three kinds of FORMABLEadaptation categories are given, as KBS Hyperbook supports adaptive information resources, navigationalstructure, trail generation and goal selection [Henze, 2000, p. 44]. Yet, other pedagogical informationabout learners has not so far been addressed. Although this system provides only a minimum of realadaptive features, the overall concept as well as its application scenario seems to be very pragmatic andutilisable.

Thirdly, [Hockemeyer and Albert, 1999] introduced the Relational Adaptive Tutoring Hypertext WWW-Environment (RATH) which was developed by Cord Hockemeyer and his colleagues at the CognitiveScience Section, Graz University. The system itself allows a learner to go through a (hypertext-based)course and complete the instructional units with concluding tasks (i.e. quizzes). As RATH is based onthe knowledge space theory, the learning progress is tracked according to the skills linked to the course’stasks. The prototype implemented at the CSS is realised with CGI-scripts within an Apache web server.Further, e-learning standards and specification are not treated in this project.

Within the scope of adaptive e-learning, the RATH system realises rather few features for adaptingthe learning process. One of the key concepts within this environment comprises the idea that achievinga task enables a learner to master a set of skills. Thus, the information about mastering certain skills isused to resolve instructional dependencies, as addressed by FORMABLE’s adaptation of the instructionalsequence. As a result, this research approach focuses, similarly like KBS Hyperbook, on content anddidactical-based adaptation, while other pedagogical states are ignored. RATH can be considered to bea prototypical research result without much applicability in practice. The strength of this solution lies inthe theoretical background, namely the knowledge space theory which is also applied in the next projectdiscussed next here.

As shown by these earlier research results, such adaptive e-learning systems mainly focus on aspects ofmacro-adaptive instructional systems, partially enhanced with intelligent tutoring systems. Nevertheless,pedagogical issues like learner characteristics or environmental states were not addressed in this area ofresearch.

6.4.2 Recent research projects

At the turn of the new millennium, new research ideas came up, most of them realised within the scope oflarge EU-funded projects. A few of them are inspected in the following paragraphs.

The first one to mention here is the so-called WebBasedTraining (WBT) Master [WBT-Master, 2007]developed at the Institute for Information Systems and Computer Media and currently applied for variouscourses at the Graz University of Technology. According to [Ebner et al., 2006], this web-based e-learningportal allows the managing of collections containing so-called Training Objects like e-books, discussionforums, chats, quizzes, virtual laboratories or project management rooms. Overall, WBT Master providesa lot of advantages like rich functionality in terms of learning activities, a high degree of flexibility, cus-tomisability and usability of the system for both learners and teachers. Technologically, WBT master isbased on web technology, whereby it is possible to apply different technologies like Servlets or PHP andalso to import SCORM packages [Helic et al., 2003].

Referring to the FORMABLE model, the WBT-Master mainly focuses on the idea of providing adapt-ability, for both learners and teachers. Concerning content model, it supports the modelling of knowledgedomains, while other pedagogical states are primarily realised by different settings in the user profile.From a didactical viewpoint, WBT-Master provides the full possibilities of FORMABLE, whether interms of so-called e-books or by applying learning goals (objectives) and learning actions (instructions).One outstanding characteristic of this system approach deals with the teacher’s possibility to adapt thelearning process utilising monitoring and tutoring tools. Thus, WBT-Master is currently restricted to pureadaptable features, although the very flexible technological and architectural framework allows the adding

6.4. Inspecting existing projects and solutions 89

of adaptive behaviour easily, for example by integrating components adapting aspects of the system auto-matically.

Another research-driven approach is the AHA! system by [DeBra and Ruiter, 2001] developed byPaul de Bra and his colleagues. AHA! comprises an adaptive hypermedia system allowing the creationand provision of hypertext-structured courses. Learners can access the content via web, while learningprogress is assessed via quizzes. The system itself is implemented as a monolithic application for Tomcat.The authoring software My Online Teacher (MOT) is available as Perl-scripts under Apache. At this stage,standardised courseware is not addressed at all.

Based on the AHAM reference model outlined in section 4.3, AHA! allows the definition of didac-tical dependencies between instructional units and, primarily, focuses on content-based adaptation. Thecourse domain is modelled with concepts, which are also used for learning tracking. Nevertheless, currentresearch activities aim to provide material for different learning styles (Kolb). Thus, the AHAM model,the AHA! system as well as authoring tools were extended in order to support pedagogical states [DeBraet al., 2002]. Yet, the adaptation towards learning styles is not assessed by observation, but only once viaa psychological test.

Amongst the younger research projects, iCLASS (Intelligent Distributed Cognitive-based Open Learn-ing System for Schools [O’Keeffe et al., 2006]) aims at developing a technological and educational frame-work to provide adaptation of learning and teaching processes. Beside adaptive functionality for thelearning in schools, aspects of creation of courseware, collaboration or reporting are also taken into con-sideration. Again, this research approach is based on the so-called “knowledge space theory”. From atechnological viewpoint, it is being realised with modern software paradigms, such as web services, learn-ing object repositories, standardised meta-data, etc. Technologically, this solution approach is based onSCORM, IMS, OWL and service-oriented standards like WSDL and SOAP.

Referring to the knowledge space theory, the iCLASS system allows classical adaptation at an in-structional macro-level by means of assessing the learner’s domain knowledge (on the basis so-calledskills), defining instructional dependencies and paths through the course, etc. Further, personalisationalso addresses new pedagogical approaches, cultural characteristics or learning styles. With respect to thethree adaptation categories of the FORMABLE model, the iCLASS prototype should provide methodsand techniques of the three adaptation categories. Overall, this research approach seems to be based ona good theoretical framework and implies a holistic and pragmatic method. Nevertheless, it cannot beexperienced yet, as it is a work in progress.

Another example of a topical solution approach is PERSO [Chorfi and Jemni, 2004] which stands for“PERSOnalizing e-learning system” and was developed at the University of Tunis El Manar. Generally,PERSO realises a typical adaptive educational hypermedia application. This solution approach mainlycomprises features to provide courseware via web and focuses only on a few real adaptation methods.PERSO is implemented as a web application within a Tomcat Server. Standardised courseware is notaddressed at all.

In short, PERSO adapts the instructional sequence on the basis of the learner’s knowledge (assessedby quizzes and open answers) and their media preferences (given via a user profile). The adaptation itselfis performed according to stereotypes and following a rule-based approach, namely case-based reasoning.In fact, the system also tries to automatically adapt the learning process by trying to retrieve scenariossimilar to former learning experiences. Aspects of pedagogically-driven adaptation are restricted to thepreferred media style, which can be set in the user profile by the learner. Although this prototype onlyimplements very few adaptive functions, it aims at adapting towards pedagogical and didactical states and,further, considers the adaptation of instructions and the instructional sequence.

Finally, the approach to Adaptive Personalised eLearning Services (APeLS) [Conlan, 2005, p. 100ff]describes another realisation of an adaptive e-learning environment. This research project deals withimplementing a distributed and services-oriented framework, where various services provide adaptive


features for e-learning. The APeLS system itself includes functionality and tools to create, manage andprovide learning content. Due to the distributed architecture, this system can be considered to be verymodular, flexible and extensible. From the technological viewpoint APeLS is based on web servicesprovided by the Tomcat server and applies XML-based specifications (IMS, LOM, etc.) for the metadata-driven adaptation approach.

The APeLS framework is a very powerful architecture for adaptive e-learning. The multi-modal andmetadata-driven approach theoretically allows all kinds of adaptation. Inspecting the actual prototype, thelearner model is mainly restricted to content-based aspects, so-called competencies and learning styles.Yet, it would be easy to extend this model and the adaptation process itself due to the service-basedarchitecture. Furthermore, APeLS implements adaptation methods of the three adaptation categories ofFORMABLE. As a conclusion, the APeLS system can be outlined as a promising framework for furtherresearch activities.

As these projects already consider pedagogical issues – at least a few ones – it remains to be examined,which impact these research results have on free and commercial e-learning platforms.

6.4.3 Open-source solutions and commercial products

As there are many e-learning solutions – [Baumgartner et al., 2002] evaluated 130 of them – only a fewwill be highlighted here, namely the free Moodle system, Cocoontec’s Campus and Studio software andthe solution of iDL Systems.

The development of the free, web-based and very popular learning management system Moodle [Moo-dle.org, 2007] was initiated by Martin Dougiamas, a former WebCT administrator at the Curtin Universityof Technology. Presently, Moodle is one of the best known and most widespread e-learning platforms inthe world and has strong community support. The popularity of this system results not only from the factthat it is completely free, Moodle provides also a lot of learning activities for distance learning phasesas well as an excellent usability for learners, teachers and administrators. Due to the large number offeatures, this LMS allows the realising of online courses with respect to all e-learning principles and evenoffers an interface for importing SCORM-compliant packages. Technologically, Moodle is based on theXAMPP (Apache web server, MySQL, PHP, Perl) packages.

Although the Moodle system does not consider adaptive e-learning explicitly, a few aspects relevant tothe FORMABLE model can be identified. First of all, all kinds of adaptation within the Moodle platformcan be reduced to the concept of adaptability by means of a user (the learner or the teacher) adapting theuser interface or the learning content. Real adaptive behaviour is not provided by this platform. Secondly,it is possible to adapt to the domain knowledge applying the “lesson” module, which provides prototypicalsequencing functions. Finally, a learner could adapt the user interface (menu, content area and layout) tohis cognitive or learning style – in fact, Moodle does not support a method to assess such pedagogicalfactors. Concerning FORMABLE, Moodle would be of interest for adaptation of learning activities, as itprovides so many modules. As this e-learning environment does not support typical features of adaptivee-learning, it is not is not relevant in context of this work.

Cocoontecs [Cocoon, 2007], a company headquartered in Vienna, develops innovative software solu-tions for human resource management and e-learning. Its product palette includes a commercial authoringtool named “Studio” as well as a learning management system called “Campus”. These two products arebased on Java web technology (Tomcat applications) and implement the standards and specifications ofSCORM and IMS. The authoring software also allows the managing of learning content distributed overvarious digital repositories. Aspects of collaboration or communication are not considered at all.

In general, these systems support content-based adaptation (the macro-adaptive instructional approach)by means of applying IMS Simple Sequencing. Further, it is also possible so create and provide learningresources for different learning styles or other characteristics, for example for a certain device or the band-

6.5. Conclusions 91

width available for a learner. The assessment of learner characteristics depends directly or indirectly onthe input of the learner, while system-driven assessment as found in ITSes is not supported at all. An ownmodel describing the adaptive behaviour of Cocoontecs e-learning platform is not available. AlthoughCocoontecs does not explicitly aim to develop an adaptive e-learning environment, a few relevant featuresfor adapting the learning process can be identified within these two products. As Studio and Campus arecommercial solutions, they deal a lot with software development, standardised courseware, usability andthe systems’ usefulness.

iDL (Intelligent Distance Learning) Systems [iDL, 2007c] is a company located in the USA. Primarily,its products aim to create and deliver online courses which adapt to the personal learning needs of eachstudent. Therefore, iDL Systems provides technological components for adaptive e-learning – such as aLMS, authoring tools or a repository – as well as services and consulting in this area. Technologically,these products are based on J2EE and support the standards and specifications of SCORM.

Concerning adaptive e-learning, iDL Systems covers a large set of features with its products begin-ning with content-based adaptation over adaptation towards learning styles up to considering self-efficacythrough intelligent feedback. Moreover, assessment of learner characteristics is not only dependent on thelearner’s input, the system also tries to retrieve information about the learner pro-actively, for example byanalysing the repetitive learning style. From a theoretical viewpoint, various journal papers [iDL, 2007a]describe the idea and research activities for the company’s products and services. Further, best practicesexperienced in different projects are published as well. As a conclusion, it can be stated that the ap-proach by iDL Systems is based on an excellent theoretical background as well as on valuable experiencesfrom former projects at various educational organisations. Moreover, the company offers a very holisticapproach for adaptive e-learning.

These three examples, amongst hundreds of existing solutions, already outline the variety of e-learningplatforms towards implementing adaptive e-learning functionality. While Moodle only provides a few as-pects of adapting the learning process manually, standard-conform products can be considered to fulfilrequirements for content-based adaptation techniques for instructions and instructional sequence. Never-theless, only a few solutions – particularly in the scope of commercial products – consider real adaptivebehaviour, as shown by the very successful iDL Systems.

6.5 Conclusions

Referring to the FORMABLE model of section 4.4, the methods of adaptive e-learning provided in liter-ature can be assigned to one of these three functional categories: (1) adaptation of the instructional level,(2) adaptation of the instructional sequence and (3) adaptation though inserting new instructions. Further,adaptive techniques also can be subsumed with these adaptation methods in order to describe systems inthe way that they can be compared to each other.

Adaptive e-learning in practice is, more or less, highly dependent on adaptable courseware, as alreadyoutlined in the last chapter. A learning platform dedicated to adapt the learning process must be able topresent the courseware and, further, utilise it, for example as adaptation information or by means of adapt-able objects. Further, an ideal adaptive e-learning environment has to provide an adaptable user interfaceand a variety of learning activities, which can be applied according to the instructions of FORMABLE.

The adaptation process itself has to be given by some specialised component which, on the otherhand, also requires modelling components to access adaptation information, systems and devices to assesspedagogical states and update these models, a rule repository as well as a trigger entity to apply theserules. The architectural design introduced in this chapter can be seen as a solid basement for realising anadaptive e-learning environment.

Finally, a short inspection of the existing palette showed that early research projects focused mainly on


content and didactical-based adaptation, while younger approaches aim to adapt on the basis of pedagog-ical states within a well-defined scope. As a conclusion, the influences on free and commercial solutionscan be identified by the fact that nearly all systems include content-based adaptation of the learning pro-cess, while the application of standards is addressed primarily by commercial products and real adaptivee-learning behaviour is implemented only by a very few companies.

Nevertheless, many prototypes resultant from research deal with adaptive e-learning, one of them isthe AdeLE system briefly introduced in the previous and described in detail in the following chapter.

Chapter 7

Technical Realisation of the AdeLE Sys-tem

“ You don’t make progress by standing on the sidelines, whimpering and complaining.You make progress by implementing ideas. ”

[ Shirley Chisholm ]

One project in the field of adaptive e-learning is AdeLE, which stands for “Adaptive e-Learning withEye-Tracking” and has been started in the year 2003 [AdeLE, 2006]. As already stated in section 5.4,AdeLE is an attempt to implement an adaptive e-learning environment based on relevant aspects of theFORMABLE model. Generally, the project aimed at two novel issues:

• On the one hand, the pedagogical model deals with exploiting eye and content-tracking technologyin order to identify the learning states of users. At this point, it should be pointed out that the useful-ness of eye-tracking technology is not examined or evaluated in this dissertation, while architecturaldesign issues resulting by the integration of such sensory systems is addressed definitely.

• On the other hand, retrieval-based adaptation is targeted by utilising the so-called Dynamic Back-ground Library [Garcia-Barrios et al., 2002]. This tool which requires a concept-based contentmodel can be of benefit for both the adaptation of the learning process and the support of teachersmanaging course materials and observing new development streams in their fields. As chapter 9introduces this system in detail, here only implementation details and benefits for online learnersare given.

Combining these two approaches into one system resulted to the AdeLE prototype, which is describedin the following section. Therefore, the planning steps of the AdeLE prototype are summarised in section7.1. Thereafter, section 7.2 and section 7.3 outline the functional units and implementation details of theprototype. Finally, section 7.4 presents a guided tour through the system and with a focus on adapting thelearning process, before this chapter as well as the practical part of the dissertation is concluded.

7.1 Planning of the AdeLE prototype

The overall development process, which was carried out by a team consisted of technologists and psy-chologists, can be divided into different stages. The first and most important one was the planning of theprototype.

93

94 7. Technical Realisation of the AdeLE System

7.1.1 Scientific approach to utilising eye-tracking technology

AdeLE particularly aimed at the application of eye-tracking technology for an enhanced observation ofthe learning behaviour. This dissertation addresses the development of an adaptive e-learning environmentallowing and considering the integration of an eye-tracking device, but does not deal with the feasibilityand usefulness of this solution approach. Although these issues are not explicitly part of this work, thissubsection gives a short overview of eye-tracking technologies and ideas about how to apply such a devicewithin the AdeLE research project.

[Gutl et al., 2005] states that the field of eye-tracking research is an old, but presently very activediscipline. Nowadays, eye and gaze-tracking systems are found in several research fields, like scan-pathtesting on the WWW [Josephson and Holmes, 2002], research in the area of aviation [Merchant, 2001] orof driving environments [Hayhoe et al., 2002], computational studies about visual cognition [Zhai, 2003],human computer interaction [Ivory and Hearst, 2001] and many others. Recently, eye-tracking vendorsbegan to implement real time eye-tracking analysis, but there is still a lack of integration with profilingsystems and exploitation of the data flow for personalised content compilation.

Therefore, the AdeLE project addressed the observation of the users’ learning activities in real-time bymonitoring behavioural aspects and personal traits [Gutl et al., 2005]. User profile information of specialinterest for AdeLE are (a) learner characteristics, such as cognitive or learning styles, (b) momentarystates, like tiredness or mental effort [Garcia-Barrios et al., 2004b], as well as (c) other indicators duringthe learning process, such as objects and areas of focus, time spent on objects, frequency of visits andsequences in which learning content is consumed [Preis and Mueller, 2003]. By interpreting this data inreal time, accurate information about the user’s state can be gained.

Exploiting eye-tracking data combined with other user behavioural traits linked with the content pro-vided, a fine-grained learner profile can be tracked by the system and applied for example for adaptationof the learning process. Based on this fine-grained learner information, it might be possible to deriveinsights into the learning strategies of users interacting with the e-learning platform and to detect pat-terns indicative of disorientation or other suboptimal learning strategies. In order to efficiently interpretbehavioural indicators, it is important to not rely exclusively on eye-tracking data, but to supplement itwith information gained by direct and constant user feedback, i.e. to achieve a scrutable adaptation of thelearning process. Further, the learner requires sufficient control over the system and the adaptation.

In terms of eye-tracking technology, eye movements, scanning patterns and pupil diameter are indica-tors of thought and mental processing involved during visual information extraction [Rayner, 1998]. Thus,real-time information of the precise position of gaze and of pupil diameter can be used for supporting andguiding learners through their learning journey. Very roughly, eye movements can be divided into twocomponents: fixations, i.e. periods of time with relatively stable eye movements where visual informa-tion is processed and saccades, which are defined as rapid eye movements that bring a new part of thevisual scene into focus. However, more important indicators can be gained by analysing both componentstogether with other derived parameters [Gutl et al., 2005].

Gaze duration (i.e. time spent on an object) and fixations are not indicative of attention per se, becauseone can also pay attention to objects that do not lie in the centre of the focused region. Nevertheless,by considering other indicators, such as saccadic velocity, blink velocity and rate as well as an eyelid’sdegree of openness, a better and more meaningful approximation can be gained. Saccadic velocity, forexample, is said to decrease with increasing tiredness and to increase with increasing task difficulty [Fritzet al., 1992]. Further, blink rate, decreasing blink velocity and decreasing degree of openness may beindicators for increasing tiredness [Galley, 2001]. Thus, if tiredness is identified, it should be possiblethrough adaptive e-learning mechanisms to suggest optimised strategies such as the best time to take abreak. At this point it is reasonable to emphasise that the user should always retain the final control overwhether to accept or reject the system’s suggestions [Gutl et al., 2005].

7.1. Planning of the AdeLE prototype 95

7.1.2 Pedagogical and technological framework

From a pedagogical viewpoint, AdeLE focussed on the following two issues: (1) Ideas derived fromthe application of eye-tracking technology and (2) didactical possibilities of utilising the Dynamic Back-ground Library.

A study to evaluate the usefulness of eye-tracking techniques in terms of scanning-paths was carriedout by the team’s psychologist. Figure 7.1 shows the device which was used for this study. As a result,[Pripfl et al., 2006] report that it is possible to detect (a) if some patterns are identifiable and usablefor differentiating among “skimming”, “reading” and “learning” activities and to detect (b) behaviouralparameters such as “reading a text”, “looking at a picture” or “looking at a navigational element”. Whilethe cognitive processes of (a) are calculated on the basis of an algorithm resultant from this study, theareas and objects focussed by the user can be resolved by functions of the eye-tracking device itself.

Figure 7.1: Utilisation of the Tobii 1750 Eye-Tracking system

Deriving requirements for the adaptive e-learning environment, the concept of learner state modellingwas introduced for the AdeLE prototype, as mentioned in [Garcia-Barrios, 2006]. Following this idea, theauthoring of learning content for AdeLE should allow a teacher to define certain didactical goals, like apassage has “to be learned” or “to be read”. With respect to section 5.4, this information can be annotatedwith the STAGE tool, whereby these goals are stored within the instructional content (as attributes ofHTML elements).

Providing the material to the learner, the LMS should be aware of such annotated elements and, further,assess if a learner achieved the targeted states. Differences between the targeted states and the learner’sactual cognitive processes should be visualised in some way. Thus, the content model and the learnermodel require some common parts in order to be able to model the didactical targets on content definedby the teacher and the actual cognitive states of the learner.

A second important issue given by pedagogy is the utilisation of the Dynamic Background Library.This tool, which will be explained more closely in chapter 9, aims at retrieval-based adaptation accordingto the learning context and the content model. Referring to FORMABLE, it is necessary that a concept-based domain model is created for a course and that different information retrieval systems can be definedfor any pair consisting of a concept and a situation. Within the learning process a dynamical generated listof queries should be provided to the learner. Each query offers background knowledge for the conceptsand the learning situations of an instruction.

From a technological viewpoint, the AdeLE prototype should consider trends and state-of-the-art fea-


tures which can be identified in the field of adaptive e-learning. Such requirements can be derived fromthe last two chapters:

• First of all, metadata should be applied as a basis for adaptation models as well as for describingadaptable objects, so that the adaptive engine can utilise this information for both the observationof the adaptation model and the adaptation of the learning process.

• Secondly, the separation and distribution of functional entities of adaptive e-learning systems issuggested by many projects and researchers in this field. Thus, a service-oriented approach shouldbe focussed for the AdeLE prototype.

• Thirdly, the eye-tracking device might continuously produce a large amount of data. Thus, efficientalgorithms and an intelligent modelling approach have to be considered.

• Fourthly, issues like a scrutable adaptation process and the overall performance of the platform areof importance, as highlighted by the evaluation study of the AdeLE prototype (see chapter 8).

The architectural design made up within the planning phase will now be summarised and furtherdecisions highlighted.

7.1.3 Overall architectural design and further decisions

The most important technological issue deals with the architectural design of the AdeLE system, as it isinfluenced by several factors. In accordance with [Modritscher et al., 2006a] and the architecture of anideal adaptive e-learning environment in section 6.3, figure 7.2 shows the overall architecture which isdivided in a server-sided and a client-sided part.

Figure 7.2: Overview of AdeLE’s architectural design

As eye-tracking devices produce a lot of data, it was decided to calculate relevant factors for the learnermodels on the client-side. Thus, the Eye-Tracking System (ETS) which tracks eye movements, scanningpatterns and pupil diameter sends this information to the Content Tracking System (CTS). Here the input

7.2. Functional units 97

stream of data is processed and features relevant to the learning process, such as the cognitive process-ing of learning content or the elements focussed by the learner are extracted. Finally, this compressedinformation is passed to the modelling system on the server side.

The server-sided part of the AdeLE prototype consists of five subsystems. Firstly, the Learning Man-agement System (LMS) should comprise all functionality of a state-of-the-art platform introduced in sec-tion 6.2. While LMS Web-Client interacts with the Eye-Tracking System, the LMS itself should access theAdaptive System (AS) as well as the Background Knowledge Repository. Models necessary for the adap-tation process are given by the Concept-Based Context Modelling System (CO2) describing the contentmodel as well as by the Modelling System (MS) managing learner profiles and models.

Early consideration of the AdeLE project included important aspects of software engineering andsystem operation, such as flexibility, extensibility, modularity, connectivity, reusability or performance.Therefore, various technologies and products like the J2EE-based JBoss Application Server, Microsoft.NET, the OSGi Service Platform and others were evaluated. In the end, it was decided to implementthe AdeLE system within the service-based framework Openwings [Openwings, 2007]. Although thisframework is rather unknown, it offers enormous possibilities for the realisation of distributed service-oriented systems.

Further decisions regarding the AdeLE approach comprise the reuse of an existing learning manage-ment system, standard-compliancy of courseware, the integration of adaptive behaviour into the system aswell as the provision of models and background information. Thus, a more detailed description of eachsubsystem is given in the next section.

7.2 Functional units

Each server-sided subsystem is intended to be one running service within the Openwings framework. Thefollowing subsections summarise functional and implementation-related details about the subsystems ofthe AdeLE prototype, whereby facts about the Learning Management System are presented first.

7.2.1 Learning Management System

In order to reuse existing e-learning software and to achieve standard-conformity, one decision in theAdeLE project was about choosing and adapting ADL’s so-called Sample Runtime Environment (Sam-pleRTE v1.3.2) [ADL, 2007b], an exemplary and free implementation of a SCORM-compliant learningplatform. Therefore, the standards and specifications had to be extended as already outlined in section 5.4.Within the architectural design, the SampleRTE stands for the LMS and its web client. From a technolog-ical viewpoint, this learning platform is a Tomcat application which allows the importing and accessingof SCORM packages.

Information about learners and courses were stored within a MS Access database. Aiming at plat-form independency, [Blaschitz, 2007] reports on the integration of the database persistence layer Torque[Apache, 2007] for the AdeLE system. As a result, the database at the system’s back-end can be ex-changed. The SampleRTE adapted for the AdeLE project is based on the MySQL database [MySQL,2007], but many other database management systems supported by Torque can be utilised very easily.

Secondly, the SampleRTE was integrated within the Tomcat server of Openwings, which allowedthe implementation of other subsystems like the adaptive engine or the modelling systems on the server-side. For the AdeLE prototype the communication between the server-sided subsystems is realised bysynchronous communication utilising Sun’s Java-based Remote Method Invocation (RMI) standard. Aspointed out in figure 7.2, the LMS invokes the API of the Adaptive System to request adaptive behaviour.Method invocations and the utilisation of the feedback given by the Adaptive System is primarily im-


plemented by adapting the SampleRTE’s so-called sequencing engine, the systemic part calculating andrendering the instructions.

On the other side, the SampleRTE also sends requests to the Background Knowledge Repository inorder to provide background information. Communication with the LMS client is bi-directional, wherebythe LMS provides the user interface and presents the courseware and the client applies the LiveConnectmethod [Mozilla, 2000] to submit the learner actions to the SampleRTE. For evaluation reasons, theseactions are logged by an own mechanism realised for the AdeLE system. Additionally, a component forform-based questionnaires and a redirector for the Dynamic Background Library were implemented andthe LMS-internal profile was extended by a learner model explained later on.

All these features are required to fulfil requirements for a controllable adaptation process as well asfor evaluation reasons as shown in the next chapter. A more detailed insight into the functionality of theAdeLE prototype is presented by a description of concrete user scenarios outlined in section 7.4. In thefollowing section, internals about the other subsystems are highlighted.

7.2.2 Adaptive System

Generally, the Adaptive System aims at adapting the learning process. The theoretical model is based onthe three adaptation methods specified by the FORMABLE model and, on a more pragmatic level, in thelast two chapters.

The models are provided by the other components, visualised in figure 7.3. While the didactical modelis defined by SCORM’s manifest file, the content model is given by the Concept-Based Context Modeller.Pedagogical states and learner characteristics are managed by the Modelling System. Mappings betweenthese models are achieved by identifiers, for example the instruction-id defines the relation between learn-ing content and the concept-based domain model. LOM’s educational attributes link learning content topedagogical states and, further, to adaptation rules. Adaptation rules and decisions are stored within adatabase.

Figure 7.3: Implementation details of the Adaptive System

The adaptation process itself considers the three important techniques implemented within so-calledadaptors and explained as follows. Firstly, adaptive sequencing of instructions comprises the idea ofadapting the path through the course on the basis of learner observation and didactic rules. Secondly,


adaptive aggregation deals with selecting and sampling resources to an instructional unit. Thirdly, adaptivepresentation describes the visualisation of an instruction as well as the navigational elements.

Yet, only a few aspects of the ideal adaptive e-learning environment were really implemented:

• The sequencing adaptor simply retrieves the sequence of instructions, without modifying them dueto a lack of an applicable model to adapt the instructional sequence within the AdeLE project.

• Adaptive aggregation of instructions focus on selecting the appropriate resource from a given list, asthere is no aggregation engine for LOM-based assets available. In addition, the aggregation adaptoralso generates the visual representation of the learner state model, which is explained in detail inthe next subsection.

• The presentation adaptor comprises any adaptation concerning navigational elements, for exam-ple the previous and next button, the tree-view navigation or the presentation of the links to thebackground knowledge.

For the prototype, the adaptation model is rather simple and generic, but highly accommodating ofthe application of eye-tracking technology. As the so-called WAVI model [Riding, 1991] was utilised,learners are described in two dimensions:

• The wholist-analyst factor (a value between -1.0 and 1.0) is calculated on the basis of the learner’susage of navigational elements. If a user prefers the previous and next button and neglects the tree-view, this factor is increased, otherwise the value is decreased. For positive values, the tree-view iscompletely disabled.

• The verbaliser-visualiser value (again between -1.0 and 1.0) is also adjusted in a very primitive way.It is increased each time a user manually selects the visual instruction and decreased on choosingthe textual information. Again, the threshold for instructional aggregation (verbaliser or visualiser)is 0.

The proprietary algorithm to adapt the WAVI-model was developed on the basis of assumptions aboututilisation of the eye-tracking device and had to work without an integrated eye-tracker, for example forthe online demonstrator of the AdeLE prototype [IICM, 2006]. The adaptation mechanism is thereforerestricted to the user interaction. Besides, the user can modify the WAVI-factors in the user profile on thebasis of a self-assessment through two understandable questions at anytime.

From the viewpoint of cognition sciences, the information for the adaptation is still restricted to theWAVI model. Nevertheless, it is easily possible to build up and use a more flexible learner model fordetermining the indicators to adapt the learning process – hence, this is no longer part of the AdeLEproject.

7.2.3 User Modelling System and client-sided sensors

The task of this subsystem is to provide and manage the user profiles as well as learner models used toadapt the learning process. According to [Gutl and Garcia-Barrios, 2005b] and [Froschl, 2005, p. 114ff],four components were identified and each one was implemented as an Openwings service (see also figure7.4):

• The Manager is responsible for communicating with external systems such as a profiler editor,the eye-tracking system, other sensory systems, an adaptive engine or even external knowledgemanagement components. It can be seen as a single and central dispatcher of the Modelling System.


• The Profiler deals with all necessary operations concerning the management of learner profiles. Inessence, a user profile consists of purpose-oriented sets of simple user attributes (e.g. static usertraits, interaction sequences, logging data, etc.), which might be accessed by other internal modulesin order to enrich its semantic value (e.g. an inference-based user model or a standard-mappingservice).

• The Modeller realises various user modelling services, which are considered as high-semantic in-formation extractors above the raw-data user profiles, for example the WAVI modeller or a dynamic(real-time) state modeller of the learner. Thus, it is unproblematic to add new modellers or moveother models like the course or adaptation model to this service of the Modelling System.

• The Data Handler provides different tools for data management (i.e. access to persistent user data).

Figure 7.4: Implementation details of the Modelling System, adapted from [Froschl, 2005, p. 118]

Overall, the AdeLE prototype is able to handle the WAVI model as well as to track dynamic real-timelearner state models. Therefore, the WAVI factors are used to adapt the learning process by means ofenabling and disabling navigational elements and presenting textual or visual instructions to a learner (seealso the previous subsection). On the other hand, the learner state model is utilised to visualise the learningprogress. Learning state modelling, as introduced by [Garcia-Barrios, 2006] and mentioned in section 5.4,deals with the idea that the teacher determines learning objectives for parts of instructions, like “a passageis to learn” or “a picture is to view”.

As the eye-tracking device can distinguish between such cognitive processing states or, at least, cancollect statistical information about the learner’s gazes, it is possible to assess these states and to givevisual feedback on the learner’s achievements on these learning states. To be more precise, the AdeLEprototype displays one of the three states to the learner:

• If the learning state is “green”, all inner-instructional objectives for this instruction were success-fully met.


• A “yellow” learning state indicates that there were no inner-instructional objectives defined or nostates assessed for the actual instruction.

• A “red” learning state means that the learner did not reach all objectives as expected. In this case,the learner can access a list displaying all elements of the instruction which were not processedproperly.

In concordance with other modern research movements, the following critical issues are consideredfor the Modelling System. Regarding scrutability and openness, the user model needs to be transparent aswell as controllable by the learner or by external applications. Mentioning access by external system, it isalso necessary to implement privacy aspects such as access control, for example by pre-defined roles andauthentication. Further, the realisation of proactive behaviour is required, so that the Modelling Systemmight force certain interactions in order to initialise or update a user model. Last but not least, a GUI-basedtool to view and edit AdeLE’s user profiles and learner models was implemented (see figure 7.5).

Figure 7.5: Graphical user interface of the Modelling System [Froschl, 2005, p. 154]

In addition to the Modelling System, two subsystems located on the client-side of the AdeLE prototypeimplement the following relevant issues within the AdeLE project. On the one hand, the Eye-TrackingSystem aims at recording the learner’s eye movements. On the other hand, the Content Tracking Systemtries to derive implications of the user’s gaze for the learning process by connecting the gaze scanningpaths with the content and updating the learner models within the Modelling System. By following thisapproach, the server hosting the LMS is not loaded by these analysis tasks, which require many resourcesto process the large amount of processing data.

As the eye-tracking device is not available for the online demonstrator of the AdeLE system, anApplet-based simulator was implemented in order to allow the learner determining the current learn-ing state for each instructional element assigned with a certain objective. Thereafter, this information


(element-id, targeted state and assessed state) is sent to the Modelling System through an Applet andutilised as described above.

7.2.4 Concept-based Context Modeller and knowledge repository

The Concept-Based Context Modelling System (CO2) described in [Safran et al., 2006] and [Safran, 2006,p. 75ff] fulfils two important requirements for the AdeLE solution approach:

• This subsystem which can be seen as the new version of the Dynamic Background Library allows themodelling of a course domain by means of concept definitions through a contextual space. Theseconcepts can be assigned to the instructions of the course and allow building a query to retrievebackground information from various repositories.

• On the other hand, the CO2 system represents a powerful additional tool, which (a) supports context-based adaptation of learning processes (i.e. thematic-driven learning as introduced in [Dreher et al.,2004b]), (b) extends the advantages of navigational elements based on the conceptual model of thecourse, or (c) enhances learning progress tracking by connecting the concepts with the learners’achievements.

Figure 7.6: Implementation details of the Concept-Based Context Modeller, adapted from [Safran,2006, p. 84]

Figure 7.6 presents the architectural design of this subsystem, which is similar to the Modelling Systemdescribed in the last subsection. The tasks of the CO2 and the resulting implementation are distinct incertain areas, however. In the CO2, the Profiler processes information about the context, i.e. as a hierarchyof context items, the concepts within this context and the matching patterns (synonyms). The Modelleraims to provide the models for a context which is retrieved from the Profiler and contains information ona higher semantic level (i.e. the current implemented models interpret the meaning of conceptual spaces

7.3. Implementation details 103

for its utilisation within the field of e-learning). The Manager and the Data Handler of the CO2 work inthe same way as the corresponding services of the Modelling System.

Within the AdeLE prototype it is possible to define and manage concepts for a course using a GUI-based tool (see figure 7.7). A teacher can therefore create a knowledge domain by defining one context-item for each instruction (even for exams), determining various information retrieval systems which pro-vide course-relevant background knowledge and connect these concepts with any context-item. Presently,the context-dependent concepts are visualised as a link-list structured by concepts and sources. Thus, theCO2 represents a didactical tool for teachers to structure their courses, to manage their static backgroundliterature and define dynamic information sources for enhancing or adapting the learning process, forexample by improved learning progress tracking, context-driven adaptation, support of thematic-drivenlearning and so forth.

Figure 7.7: Graphical user interface of the Concept-Based Context Modeller [Safran, 2006, p. 106]

The Background Knowledge Repository comprises all information retrieval systems (IRS) addressedby the CO2. Since arbitrary IRSes can be defined as knowledge sources for the concepts, it is possibleto directly access an internal IRS or even an external one by implementing its query definitions or API.For example, a teacher could use Wikipedia [Wikipedia, 2007] to provide the definition of a concept,Google Images [Google, 2007] to retrieve concept-related pictures and online dictionary services to offertranslations. The IRS currently integrated into the AdeLE prototype is the xFIND system [Gutl, 2002,p. 255ff], a smart knowledge discovery system developed at IICM.

7.3 Implementation details

Right after this comprehensive overview of AdeLE’s functional units, a few implementation details arepointed out in this section.


7.3.1 The Openwings framework

In order to find a general architectural solution for the AdeLE research project, some basic technical re-quirements have to be stipulated and evaluated. As a result, the idea behind the conceptual design of thetechnical architecture of AdeLE is based on the following software system requirements, such as an easyextensibility, open interfaces, strict modularity, high scalability, the atomicity of software components,encapsulation of different scopes of functionality, the ability to integrate network functionality, exchange-ability of components as well as the utilisation and integration of standards.

The key issue in the context of the AdeLE project was determined by the requirement for atomicsoftware components, which defines the smallest software unit able to provide the required functionality tosolve a specific problem. Existing technologies refer to such specialised units as, for example, “services”,“components” or “agents”. As a result of inspecting technologies and products like the J2EE-based JBossApplication Server, Microsoft .NET or the OSGi Service Platform, as well as of examining researchprojects like KOD [Sampson et al., 2002a], SeLeNe [SeLeNe, 2007] or AEHS [Dolog and Henze, 2003],the Openwings framework was applied for the research project.

Figure 7.8: Top of the Openwings Explorer displaying installed components

After installing and starting Openwings, the so-called Openwings Explorer appears on the screen. Thismonitoring and management tool provides an overview of all installed components and interfaces as wellas the running services. In the case of the AdeLE prototype, figure 7.8 lists all Openwings-specific andAdeLE-related components, whereby the ones with the prefix “AMS ” comprise the Modelling System,the ones with the prefix “DBL ” stands for the Concept-Based Context Modeller and the adaptation-


provider (in combination with the configuration-provider) represents the Adaptive System. The LearningManagement System, namely ADL’s SampleRTE, is installed the Openwings-intern Tomcat server.

Figure 7.9: Bottom of the Openwings Explorer displaying interfaces and services

In figure 7.9 the interfaces between the installed components as well as the running services are shown.The communication between the services is realised with the RMI standard applying the interface compo-nents for the service lookup. The Openwings Explorer itself provides necessary functions to administratethe platform and to install or remove, to configure and to start components. Further, this tool is also helpfulfor monitoring running services, for example by functionality like showing a log console for each process.Finally, Openwings also allows system administrators to move running processes to other Openwingsinstances.

As all relevant services are started, the AdeLE prototype is available via web browser. A concretescenario of learners interacting with the system is shown in the next section.

7.3.2 Implementation of the Adaptive System

As the Adaptive System constitutes the heart of the adaptive e-learning environment, the implementationof this system is of outstanding importance, not only for the usefulness of adapting the learning process,also for the usability of the overall system. The approach towards AdeLE’s Adaptive System is based ondifferent aspects, partially resulted from evaluation studies coming up in the third part of this dissertation.

Firstly, the dependency of this system from various models is considered in the following ways:

• As the didactical model can be seen as the most important one, it is received from the LMS bymeans of the XML-based SCORM manifest file. For performance reasons, the DOM representationis kept in memory, as long as at least on learning session requires a didactical model.


• The concept-based domain model is also retrieved on the first time it is needed. Yet, this model isprovided by the Concept-Based Context Modeller, so the Adaptive System can request this modelpro-actively. As courses imported into the LMS cannot be modified later on, the content model canbe considered to be unchanged at least over a learning session. Again, this model is retrieved onfirst request and cached in memory until the Adaptive System is stopped.

• Learner models and profiles are requested from the Modelling System. Hereby, no caching is im-plemented as it is important to access accurate information about the learner. Learner models areutilised for the adaptation process, while profile information can be also delivered to and from theLMS, for example if a learner registers or modifies the user profile. The communication betweenthe LMS and the Modelling System is always mediated via the Adaptive System.

• The adaptation rules are saved within a database. In a first step, didactical rules are extracted fromthe course model given by the SCORM manifest file. This rule specifies which aspects of the courseare adaptable. For instance, adaptive aggregation requires the didactical rule that there is a visualand a textual resource for one instruction. In the second step, a learner model is applied to determinethis rule, for example in order to personalise the instruction on the basis of the verbaliser-imagerfactor mentioned in the last section.

• If the rule cannot be fully applied towards a learner, a form-based questionnaire is generated. Utilis-ing this form, the learner can give feedback in order to accomplish an adaptation decision. Further,the presentation adaptor also provides some kind of information about the adaptation process aswell as control elements to overwrite the system’s decision. In this way, the adaptive behaviour ismore scrutable and always controllable by the user.

From the software development viewpoint, the adaptation techniques are realised by the adaptors,which are implemented according to the Product Factory design pattern combined with a Singleton ap-proach. To give an example, the scenario of a learner entering a course is used. If the learner launchesa course, three adaptors – a presentation adaptor, an aggregation adaptor and a sequencing adaptor – arerequested from the product factors. Yet, these adaptors are instantiated as singletons, so they are kept inmemory until the learner quits the course. While the approach with the product factory fulfils the require-ments of the AdeLE prototype, it would be better to apply an Abstract Factory to handle a product familyof adaptors, for example for a different e-learning standard or platform.

Furthermore, the adaptation trigger is implemented with the adaptation-provider, the Manager entityin figure 7.3. This service requests the three types of adaptors and executes the methods of a genericadaptor: (1) gathering the course model which is cached after the first request, (2) adapting the chosenaspect (presentation, aggregation or sequence) by the didactical rule, (3) adapting by the learner rule(personalisation step), (4) adapting by learner feedback if required and (5) creating an LMS response.

To conclude the implementation details about the Adaptive System, it has to be mentioned here thatthis subsystem consists of two processes, the adaptation-provider and the configuration-provider. Thesecond process deals mainly with low-level systemic tasks, like configuration management, the data-layerincluding access to the database as well as different XML-specific operations. These functions wereencapsulated to an own service, because they could be necessary for other systems, too. Overall, theAdaptive System has a strong focus on performance issues, as noticeable for example by the caching ofmodels of the application of the Singleton pattern. This very special software requirement can be seen asone special result of an evaluation study depicted in chapter 8.

7.3.3 Interaction with other systems

Drawing conclusions from the last subsection, the Adaptive System can be considered to be a very centraland important component of the overall adaptive e-learning environment. Thus, it has to interact with


various other systems. As communication is realised asynchronously, system interaction can be dividedinto two categories: On the one hand, the Adaptive System provides certain services for other components.On the other hand, it utilises RMI methods provided by other services.

Within the context of the AdeLE project, the Adaptive System was mainly implemented for the pur-poses of this project. Contrary to a multi-purpose adaptive engine described in section 2.4, this systemis only applied by the LMS. Precisely, the SampleRTE adapted for the AdeLE approach invokes a noti-fication service of the Adaptive System, which returns a feedback object containing a command as wellas information related to the answer. For instance, the feedback could comprise the decision to adapt aninstruction and include the identifier of the resource to be displayed.

Figure 7.10: Sequence diagram for the scenario “learner navigates instruction”

Another example deals with requesting the SCORM manifest file which is applied as a didacticalmodel below. Therefore, a special command encapsulated in the feedback object is returned from theAdaptive System to the LMS at the first time this XML-based file is required (“lazy retrieval”). As a result,this XML-based file is sent to the Adaptive System and stored within its database in order to guaranteefast access to it and prohibit unnecessary network traffic. Overall, this request-feedback mechanism wasnecessary to enhance the synchronous method invocation between the LMS and the Adaptive System bysome kind of bidirectional communication. However, it was necessary to extend the LMS with businesslogic to understand and react to the feedback objects.

The second category of interaction between the Adaptive System and other systems comprises the


utilisation of other services, which is supported by the Openwings framework. To encapsulate the way toaccess the services provided by other systems, an own Adapter class was implemented for each subsystem.As the interaction always addresses the same systems, the Adapter objects are dedicated to the Singletonpattern. Again, they follow the principle of “lazy creation”, so they are instantiated as soon as one of theadaptors requires them for the first time.

To give an example of the systems’ interaction between themselves, figure 7.10 illustrates what hap-pens within the prototype if a user navigates to a certain instruction. As the LMS receives this learneraction, it notifies the Adaptive System in order to start the adaptation process. On the first request, theAdaptive System has to prepare all necessary models, like the didactical model – the content model givenby the Concept-Based Context Modeller is omitted to simplify this example. The Adaptive System thenhas to apply the didactical and pedagogical rules in order to personalise the instruction towards the learner.If this is not possible, a form is generated and presented to the learner, who can finalise the adaptation de-cision and receive the instruction on the screen.

This rather simple use case already demonstrates that a lot of systemic interaction occurs in the back-ground, which already outlines the importance of performance issues like caching the didactical or thecontent model, applying the Singleton pattern to avoid the system’s resources, etc.

7.3.4 Adaptation methods and techniques of the prototype

At the end of this section the adaptation methods and techniques realised in the AdeLE prototype are sum-marised. With respect to the FORMABLE model, the following adaptation methods have been consideredby the AdeLE system so far:

• First of all, adaptive content aggregation is partially implemented. On the one hand, each instructionis assembled by selecting appropriate resources included in the SCORM packages. The decision isdrawn on the basis of the mapping between the verbaliser-imager factor of the learner’s WAVI modeland the learning resource type of the files which are described by the LOM standard. On the otherhand, the presentation adaptor adjusts navigational elements and provides information about theadaptation process for scrutability reasons. Hereby, the tree-view navigation is enabled or disabledaccording to the wholist-analyst factor of the learner model. Further, links to alternative instructionsas well as the learning state model is displayed within an own area on the screen.

• Secondly, adaptive instructional sequencing is at least considered by an own adaptor. As no con-crete requirement for adapting the instructional sequence was given, the sequencing adaptor extractsthe instructional sequence specified within the SCORM manifest file and encapsulates it. Thus, itwould be easy to adapt this sequence by some kind of pedagogical model. Sequencing rules them-selves are part of the SCORM-based course and interpreted and realised within the LMS. From theviewpoint of the didactical model, this method is fully implemented, while the AdeLE system lacksa pedagogical strategy to adapt the instructional sequence.

• Thirdly, the Dynamic Background Library described in chapter 9 deals with adaptation of coursesthrough a retrieval-based approach. Consisting of the Concept-Based Context Modeller and theBackground Knowledge Repository, this solution allows the definition of queries for instructionalmaterial related to a certain context or situation. In the current implementation of the adaptation pro-totype, this background information is simply presented as is and it is not added to the instructionalsequence. Yet, the link-list providing the background knowledge could be adapted on the basis ofdidactical rules or even pedagogical states, which is going to be explained later in this work.

7.4. A walk through the AdeLE system 109

Addressing adaptation techniques, the AdeLE prototype realises a few of them. According to the threecategories of adaptation methods, the following statements about such techniques can be manifested here:

• Adaptation of instructional level is realised, for example, by providing instructions on different lev-els of detail and in a multimodal way as well as adapting navigational elements. Due to the fact thatAdeLE lacks an aggregation engine for instructions, fine-granular techniques like conditional textor fragment variants are not considered at all. Yet, the adaptation towards visualisers and imagers isequivalent to page variants. The manipulation of the tree-view navigation can be considered to bea raw way of link hiding. Further, the highlighting of the actual instruction in the tree-view, whichwas implemented within the project, deals with the link annotation technique.

• Adaptation of the instructional sequence is, at least, realised from a didactical viewpoint by meansof the sequencing rules given by SCORM. Hereby, adapting the underlying function of the nextbutton can be seen as the most important element of this category. For instance, the next buttoncould be utilised to lead to the most relevant instruction or to leave out certain instructions.

• Finally, the Dynamic Background Library aims at adapting a course through inserting new instruc-tions. Although these additional instructions, which might be offered for different pedagogicalreasons, are not persistent in the course structure, this approach allows the retrieval of new coursematerials on the basis of dynamically generated query-links. As a result, the original course speci-fied with SCORM can be enriched with different types of learning content and activities.

Overall, the AdeLE solution approach considers the classical adaptation methods adequately and,further, implements a selected set of techniques. To give an insight into this prototypical adaptive e-learning environment, the following section describes the user-view in terms of concrete learner use cases.

7.4 A walk through the AdeLE system

From the viewpoint of user-centric design, two primary roles can be identified for this system, the learnerand the teacher. In the following three subsections the learner-centred functionality is highlighted, wherebyeach of these features can be experienced by the online demonstrator of the AdeLE prototype under the ad-dress http://adeledemo.iicm.edu:8880/adl [IICM, 2006]. Thereafter, administrative functions for teachersare explained, before this chapter as well as the practical part of the dissertation is concluded.


7.4.1 Registration and login

Before one can login to the AdeLE online demo, one has to register to the system. Figure 7.11 shows thelogin dialog of AdeLE, which was adapted from ADL’s SampleRTE by adding a “Register” button.

Figure 7.11: AdeLE system login dialog

Pressing this button, the user is redirected to the registration form presented in figure 7.12. Here, afew fields, like the username, the first and the surname as well as the password, have to be entered, if onewants to get access to the online demonstrator.

Figure 7.12: Registration form


After submitting this data, the user profile is created and the account is immediately valid. Thereafter,a user can log in to the AdeLE system using the username and the password.

Figure 7.13: AdeLE prototype main menu

After successful authentication, the main menu of the adaptive e-learning environment appears (seefigure 7.13). At this point, the user can access typical learner-specific features outlined in the next sub-section, enter one of the own courses as explained in the subsection after the next one or simply log offagain.

7.4.2 Learner-specific features

Learner-specific features comprise course enrolment, the modification of one’s profile as well as viewingown learning progress.

Figure 7.14: Dialog for course enrolment


Following the “Register for a Course” link in the main menu opens the course registration shown infigure 7.14. Here, the learner can enrol or unregister from each course available in the platform by simplyselecting or deselecting each course and submitting this information. Thereafter, the user returns to themain menu.

Figure 7.15: Overview of the learning progress for an example course

Another feature of the main menu which can be reached by the “View Course Status” link gives anoverview of the learner’s achievements in each course. Therefore, the user has to select one of the coursesand receives information about the learning progress in this course, as presented in figure 7.15.

Figure 7.16: Form-based dialog to edit the user profile

Finally, a learner can also modify their profile by clicking on the link “Change My Profile”. Usingthe dialog shown in figure 7.16, profile information like the password or various attributes describing thelearning style and preferences can be entered. Additionally to these fields which are part of the Sam-pleRTE, the user profile was extended by the WAVI model as outlined in the previous sections.


Therefore, a learner can self-assess both the wholist-analyst factors and the verbaliser-imager factorin terms of two statements and by using the radio buttons. This information, which is stored within theModelling System, is utilised to adapt the learning process, as already mentioned in former sections andillustrated in the following.

7.4.3 Navigating a course

In the main menu, a learner can enter a course by clicking on the link “View Registered Courses” andchoosing one of the courses. Thereafter, the selected course opens up, as for example shown in figure7.17.

Figure 7.17: Learner’s view of a course

This view of a course consists of three areas: Firstly, the header displays information about the LMSas well as the buttons necessary for a learner. While the buttons “Suspend”, “Quit”, “Continue” and“Previous” were given by the SampleRTE, the button “Simulate Eye-Tracker” was added for the AdeLEapproach, as explained later. Secondly, the left side includes further navigational elements like the tree-view, the link-list for background knowledge (not available in this figure) as well as information aboutthe adaptation process and the learner model. Thirdly, the content area on the right side displays theinstructional content.

At this point, the learner can interact with the system in the following ways: On the one hand, userscan navigate through the course, whether by consuming the instructions stepwise using the previous andcontinue buttons or by visiting randomly via tree-view navigation, as far as instructions are accessible(clickable) in the tree. On the other hand, learners can consume the instructional content, which might be


an HTML page or any other format compatible with the browser. Gaze tracking works only for HTML-based content due to technical restrictions of the eye-tracking device, however.

Figure 7.18: Example for an examination including an assignment task

Beside the passive cognitive processing modes, instructions can also comprise examinations, for ex-ample like the one presented in Figure 7.18. The utilisation of such quizzes allows the assessment ofthe learning progress on the basis of so-called “Objectives” which can be defined in SCORM-compliantcourseware. As already pointed out in the last subsection, learners can access information about theirprogress with respect to these objectives, while teachers can even manipulate them, for example by reset-ting. Nevertheless, these objectives only describe a certain learner state and include no information abouta type or a level, so they are not comparable with the learning objectives outlined in section 3.3.

Figure 7.19: Tree-view navigation of the AdeLE system


For the AdeLE system, the tree-view shown in figure 7.19 was slightly enhanced in terms highlightingthe current instruction. Additionally, this navigational element is adapted within the virtual learning and onthe basis of the wholist-analyst facts, as mentioned before in this chapter. If this factor is adapted towardsthe analyst extreme, the current adaptation model assumes that the learner does not require the elementsin the navigation area and simply hides them. In order to provide a controllable adaptation process, thenavigational elements can be enabled again by the control element displayed in figure 7.20.

Figure 7.20: View of the navigation area with hidden elements

Extending the original LMS by aspects of the AdeLE approach, the “Background Knowledge” andthe “Why this way?” sections were added. The visual representation of the background knowledge, aspresented in figure 7.21, consists of a list which is structured by the concepts assigned to the currentinstruction. For each concept, several queries for different information retrieval systems might be defined.

Figure 7.21: “Background Knowledge” section for an exemplary instruction


The section “Why this way?” as shown in figure 7.22 displays information about the adaptation processand about the learner model. Thus, shortcuts to alternatives to the adaptive decision are listed in order toprovide an understandable and controllable adaptation. Further, the two WAVI-factors (wholist-analyseand verbaliser-imager), as well as the state model of the current instruction, are presented to the learner.

Figure 7.22: “Why this way?” section for an example learner

As mentioned earlier in this chapter, a yellow learning state means that there are no inner-instructionallearning objectives defined or no cognitive processing states can be assessed for this instruction. A green-coloured state would indicate that the learner achieved all didactical objectives for the instruction, whilethe colour red informs of missed targets.

Figure 7.23: Form-based eye-tracking simulator

The final AdeLE prototype assesses the learner state model by applying the eye-tracking device. Asthe web-based online demonstrator lacks such hardware, which has to be fully integrated into the platform,a form-based simulator was realised (see also figure 7.23). This web application analyses the instructionalcontent for the tags defining the learning objectives and presents them as a table. Now, the learner cansimulate the functionality of the eye-tracking device by telling the AdeLE system, which learning stateswere achieved for each tagged element. After submitting this information to the Modelling System viaApplet, the “Why this way?” section is updated to visualise changes of the learning state.

In addition to all these functionalities of the original LMS and the extensions through the AdeLEproject, two further components were added: The first one comprises a tool which presents an HTML-based form to the learner and collects the information submitted within a database. The second componentis a logging mechanism within the LMS, which was necessary to track the learner behaviour more exactly.Both components were necessary for example for the evaluation study pointed out in the next chapter.


7.4.4 Administrative features for teachers

In accordance with the SampleRTE, teachers can access the system in the same way as learners, buthave more possibilities within the platform. Because it is not possible to sign up as teacher, anotheradministrator has to set a special flag for instructors. As a result, an authenticated teacher receives severaladditional functions in the main menu. An overview of these possibilities is given in figure 7.24.

Figure 7.24: Menu with additional functions for teachers

In detail, the following administrative tasks can be achieved by teachers within the AdeLE prototype:

• Course management deals with managing the SCORM packages. Overall, a teacher can import(“Import Course”), modify (“Manage Courses”) and delete course packages (“Delete Course”)within the platform.

• Observation of the learning progress can be achieved via “View All User’s Course Status”. Here, ateacher can access the states of the objectives for all learners. These objectives have to be determinedwithin the SCORM packages.

• User management comprises all activities to create, edit or delete learners. Therefore, the links “AddUsers”, “Manage users” and “Delete Users” offer adequate functionality. In certain situations thelinks “Global Objectives Administration” and “Clear Database” might be useful to reset objectivesfor selected courses or clear the whole database.

Beside these administrative tasks within the AdeLE platform, further activities have to be accom-plished in order to provide an adaptive e-learning course. More precisely, these steps are necessary tocreate an AdeLE-conformant course package:

• First of all, the learning content has to be determined, whether by creating it from scratch or byreusing existing material. Hereby, it might be necessary to convert content into HTML format toenable the options of the eye-tracking device. For creating quizzes, it is recommended to applyIMS-QTI-compliant templates or tools.

• To create the SCORM package, the XML-based manifest file including the organisation, the se-quencing, objectives, the content packaging, etc. has to be generated, whether by hand or by util-ising a tool. Each resource file of the learning material has to be described by the educationalattributes of LOM, as this information is required for automatic adaptation of instructions. Withinthe scope of the AdeLE project, the open source Reload editor [Reload, 2007] was applied for thistask.


• Thereafter, The SCORM manifest file has to be enhanced manually. With respect to the standard-based approach to adaptive e-learning (see section 5.4), each instruction can be extended withfurther resource files, again described with certain LOM attributes. On the basis of this meta-information, the aggregation-adaptor selects the appropriate resource for a learner. If this is notpossible, learner feedback is required.

• For the state modelling approach, it is necessary to tag elements of instructional content using theSTAGE tool introduced in section 5.4. Tagged elements are utilised by the eye-tracker or the eye-tracking simulator application in order to update the learner’s state model.

• Finally, the concept-based domain model and the queries for the background knowledge repositorieshave to be created and assigned to the course. Thus, a teacher might apply the GUI-based applicationoutlined in the last section. As this tool is rather blunt, the domain models and queries for theexisting AdeLE courses were built up using XML editors.

At this point it has to be stated that the effort required to create adaptable courseware for the AdeLEsystem is rather high. Generally, this aspect is one of the most significant disadvantages of adaptive e-learning, as also highlighted by [Brusilovsky, 2003b]. However, considerations towards the creation ofadaptable courseware are not part of this work.

7.5 Conclusions

Concluding this chapter, as well as the practical part of the dissertation, the FORMABLE model seems tobe applicable in practice, even if used in a rather informal way as done in the last three chapters. On theone hand, this formal model was applied to identify and manifest requirements for adaptable, standardisedcourseware. Consequently, existing specifications and standards in the field of e-learning were inspectedin order to support adaptive e-learning. Further, a standard-based approach for the AdeLE research projectwas introduced.

On the other hand, FORMABLE was also utilised for categorising methods and techniques of adaptivee-learning. Furthermore, an ideal model of an adaptive e-learning environment was described in terms ofconcrete requirements for such a system. After an inspection of selected research activities, projectsand products, the technical realisation of AdeLE was summarised. Hereby, aspects of standard-based,adaptable courseware and of the ideal, adaptive e-learning system were taken into consideration.

As pointed out in this chapter, the AdeLE prototype was planned and realised on the basis of theFORMABLE model and along certain requirements given by research objectives. The two most importantrequirements comprise the application of eye-tracking technology within the system and the utilisation ofthe Dynamic Background Library to achieve retrieval-based instruction. These two prerequisites hada deep impact on technical issues like the architectural design, the necessity for components like theConcept-Based Context Modeller, implementation details of the Adaptive System and so forth. But theauthoring of courseware was also influenced in the way that the effort for creating AdeLE-compliantcourses is more complicated and costly.

The service-oriented architecture of the Openwings reference implementation supports the approachto understand e-learning as a tool repository and allows the easy integration of further tools. Further-more, software paradigms such as flexibility or scalability are considered and supported, for example bydeveloping new services or relocating heavily loaded services onto other Openwings platforms. Overall,the prototype implemented within the scope of the project can even be experienced in terms of an onlinedemonstrator.

7.5. Conclusions 119

Nevertheless, the usefulness and the usability of the AdeLE solution approach needs to be evaluated.Therefore, the third part of this dissertation comprises a proof of concept for the prototype itself, as wellas for a tool realised for and ideas related to this research project.

III. Proof of Concept

121

Chapter 8

Adaptation of the Learning Process with-in the AdeLE Prototype

“ Nothing can have value without being an object of utility. ”

[ Karl Marx ]

In the recent chapters aspects of adaptation systems, technology-based learning and teaching and adap-tive e-learning are examined from a theoretical viewpoint and the development of an adaptive e-learningplatform, the AdeLE prototype, is described. Hence, the usefulness and usability of this system have notbeen examined so far. Therefore both the current and the next two chapters deal with the evaluation of theAdeLE system itself, a tool adapted for and applied within the prototype and the overall idea of adaptingthe teaching strategy.

Starting with the adaptation model realised for the AdeLE project, this chapter reports on a studycarried out at the Technical University of Graz which aimed to evaluate the prototype. Thus, section8.1 summarises the planning stage by outlining the setup of the study, the former implementation of theprototype as well as the evaluation method and the outcome expected. Thereafter, section 8.2 highlightsthe experiences gained from this study, before section 8.3 presents evaluation results of other researchers.

8.1 Planning stage

After three years of research and development, it was planned to evaluate the actual implementation of theAdeLE prototype in order to evaluate whether the system is utilisable and useable within the context ofe-learning.

8.1.1 Setup of the study

Concerning usefulness and usability of the prototype, a case study was conducted in February 2006. Over-all, 60 students were instructed to participate in an online course and complete it by passing three onlinequizzes. Further, they had to fill out a pre and a post-questionnaire, for example for a self-assessment oftheir WAVI-factors and give feedback about the system’s usability. The students also had to complete theso-called VICS v2.2b program, a psychological test to retrieve the WAVI-factors [Peterson et al., 2003],at the end of the online learning session. In addition, the students’ interaction with the system was trackedvery carefully and conclusions were drawn from these log files.

123

124 8. Adaptation of the Learning Process within the AdeLE Prototype

As the laboratory consisted of only 22 workstations, the whole class was divided into three groups of20 students each and the study was conducted in three stages. Furthermore, in each run the 20 studentswere further divided into smaller subgroups in order to examine the adaptation model, as explained later.The AdeLE system was installed on a Tomcat server available over the Internet so that the students couldaccess the learning platform from each workstation with Internet Explorer. Additionally, the VICS v2.2bprogramme was pre-installed on each computer.

The online course created for this study deals with the topic of “adaptive e-learning technologies”and is divided into three modules: The first one instructs learners about theoretical issues of adaptivee-learning, like adaptivity and adaptability, e-learning and learner-centred adaptivity in the field of on-line learning. The second module presents technological components like learning management systems,adaptive engines and modelling systems to the learners. The third one comprises further aspects of adap-tive e-learning, such as assessment, domain modelling and the provision of background knowledge.

Each module has to be finalised with an online examination consisting of quizzes with different ques-tion types, for example multiple choices, assignment tasks and short answers. The students’ input isevaluated and immediately sent to the platform. Accurate feedback is given in terms of failure statistics.In order to successfully complete a module, at least 50% of the questions have to be answered correctly.Additionally, the second and the third modules are accessible only by successful completion of the firstpart of the course (didactical dependency). For each instruction (except for tests) a conceptual modelexists and adequate background information from different information retrieval systems, like Google,Google Images, LEO dictionary, Wikipedia, etc., are provided.

On entering the course the student is instructed to fill out a pre-questionnaire by following a link.Thereafter, a form containing questions seeking general information about the student is opened within anew window. After submitting this data, this window is closed again and the student can start the learningprocess. Reaching the final instruction, which is only available after all tests have been completed suc-cessfully, two further evaluation tasks are presented. The first one leads to a post-questionnaire consistingof questions on the usefulness and usability of the system. The second one instructs the student to startthe VICS v2.2 tool and complete the psychological test. As a result, the WAVI factors are acquired by thisprogram.

8.1.2 Former prototype and adaptation model

At this time, the AdeLE prototype was not fully implemented as described in the last chapter. The follow-ing features were provided by the system:

• The core functionality for learners is essentially the same as presented in section 7.4. The maindifferences concern missing functions in the background, implementation details as well as a fewvisual elements within the course delivery engine.

• First of all, information about the adaptation process and the user profile was not displayed inthe navigation area. The system provided elements which allow the learner to select alternativeinstructions, however.

• Second, the adaptation of the tree-view could not be controlled at all. If the adaptive engine decidedto disable this navigational element, the user has to suspend the course and edit the wholist-analystfactor in the user profile.

• Third, caching of the concept-based domain model and the queries to the background informationwas not implemented within the Adaptive System so far. This aspect had a severe impact on thesystem’s usability as stated in the next section.

8.1. Planning stage 125

• Fourth, the functionality of the Dynamic Background Library was disadvantageous in terms oftracking user behaviour. As the background information was opened within a new browser window,learners were not forced to close this window again.

• Fifth, the learning state models were not considered due to the fact that the eye-tracking device wasnot yet integrated and the simulator would not be applicable for a real e-learning phase.

Beside these differences to the recent implementation of the AdeLE prototype, the adaptation modelwas not also completely up to date. On the one hand, the algorithm adapting the WAVI factors was imple-mented rather prototypical. Concretely, the two factors of this model (both values ranging from -1 to 1)were adapted by steps of 0.1 on certain user actions. For instance, if a learner chose an instructional alter-native, which indicates that one overrides the adaptation decision, the verbalise-imager factor is increasedor decreased by 0.1. Similarly, the wholist-analyst factor is adjusted on the basis of the navigationalelements used by the learner.

On the other hand, learning state modelling was not evaluated within this study. Although this is animportant issue for the AdeLE approach, it would have made no sense to examine this kind of adaptationdue to the fact that 20 eye-tracking devices would have been necessary and, further, the application ofthe simulator application would not be very reliable. As a result, the adaptation model for this study wasrestricted to didactical rules, the provision of background knowledge as well as adaptive presentation andinstructional aggregation on the basis of the WAVI model.

8.1.3 Evaluation method and expected outcome

The study set up in February 2006 at the Graz University of Technology aimed at evaluating the followingaspects:

1. The adaptation model of the AdeLE prototype should be examined in different ways and by dif-ferent factors: On the one hand, the 20 students of each run were divided into 4 subgroups of 5students. Each of these subgroups started with a pre-initialised WAVI model (wholist-verbaliser,wholist-imager, analyst-verbaliser, analyst-imager) to draw conclusions about the impact of thesestart values on the adaptation model. On the other hand, the WAVI-factors gathered from threesources were analysed: (a) the values of the WAVI-models after a student completed the course, (b)the factors self-assessed by the students via post-questionnaire and (c) the results of the psycholog-ical test implemented by the VICS v2.2 program. Further, the history of the WAVI-factors can bereconstructed by querying the log entries within the database.

2. This study also aimed to evaluate the idea of the Dynamic Background Library by inspecting if andhow the background information provided by the AdeLE system was used.

3. The students’ learning behaviour and didactical outcomes were intended to be examined on the basisof the user actions logged by the LMS and the information collected by the Modelling System.

4. Finally, the quality of learning material as well as the usability of the system is analysed on the basisof the students’ feedback given by the post-questionnaire.

As the first two issues address the usefulness of the AdeLE prototype, it was expected that it is possibleto realise online courses with the AdeLE system. Teachers estimated the duration of the learning phase tobe about 1.5 hours, in which all students should be able to complete the course successfully. Concerninglearning behaviour, a lot of statistical data like the number of clicks, the average time for each instruction,the navigational elements primarily used, etc. might be collected.


Furthermore, the examination of the adaptation model should lead to a lot of evaluation data, i.e.concrete values for the two WAVI-factors from different sources. Yet, it was not predictable, if and howstudents would benefit from the adaptation of the online learning process. While the usage of the DynamicBackground Library and the analysis of the learning behaviour would also deal with log data like clickbehaviour, aspects of the quality of the course and the system’s usability would comprise feedback givenby the students and be interpreted by the evaluators manually.

8.2 Experiences gained

Running the case study leads to lots of evaluation results and to various systemic improvements. Interest-ing observations and influences on the AdeLE system could be manifested not only after the study ended,but also within the implementation stage.

8.2.1 Implementation of the study

The study was set into practice as planned. Nevertheless, during the course of the three runs the followingproblematic aspects occurred:

• A few days before the first run started the VICS tool was delivered. Yet, the psychological testseemed to be inadequate for the students, as it was in English and a lot of very specific terms wereused. As the understanding of these words is of prime importance for the test, evaluators had totranslate the language resources into German.

• Dividing the 60 students into four groups and pre-assigning each group initial WAVI-factors viainstruction, did not work fully. One of the students selected the wrong value in the user profile,although the WAVI model was initialised at least in the correct quadrant. Thus, this error can beignored.

• Right after starting the first run of the study, severe performance problems occurred due to im-plementation failures in one of the subsystems. Consequently, this usability issue had a negativeimpact on learning. Before the second pass was started, the problem could be identified within theConcept-Based Context Modeller, where the domain model (a large XML file) had to be parsedon each request. Thus, if the AdeLE system requests the links to the background information, theserver was under a heavy load and students had to wait for up to one minute to request instruction.

As a quick solution, a caching strategy for the domain model which cannot be modified after being im-ported into the system was implemented within the Adaptive System (see also section 7.3). Consequently,the system’s performance could be improved slightly for the second run of the study and clearly better forthe third one. Although the case study did not fully run as planned, interesting findings on the system’susefulness and, particularly, on its usability could be gained, as summed up in the following subsections.

8.2.2 Usefulness of the system

Concluding from the case study, the usefulness of the AdeLE prototype is examined from three viewpoints,the students’ learning behaviour, the applicability of the adaptation model as well as the utilisation ofthe Dynamic Background Library. Addressing the adaptation techniques realised in the platform andapplication of the WAVI-model as the basis of the adaptation, the WAVI-factors were gathered from threesources: (1) the VICS test, (2) the students’ self-assessment and (3) the results of the AdeLE system usage.

8.2. Experiences gained 127

Figure 8.1: Distribution of WAVI-factors given by the VICS tool (yellow triangles), students’ self-assessment (green diamonds) and the AdeLE system after completing the course (red squares)

The distribution of the factors is visualised in figure 8.1, whereby the factors of the VICS tool had to betransformed to the AdeLE-internal representation.

With respect to the reliability of the VICS programme [Peterson et al., 2003], this psychological testmanifested the ratio between wholist and analysts as 6 to 54 (0.11) and the ratio between verbaliser andimager as 22 to 38 (0.58). This significant profile of the WAVI-factors could be an indicator of a ratherhomogeneous student group of a certain domain. In accordance with [Phillips, 2005], it can be stated thatindividuals are (usually) not good judges of their own learning styles. In the case of this study, a lot ofstudents considered themselves to be analysts and imagers, though, in this connection, an exaggerationconcerning the intensity of the learning style could be recognised.

Characteristic Analyst- Analyst- Wholist- Wholist-Verbaliser Imager Verbaliser Imager

1. Avg. number of clicks per student 30.8 28.3 26.3 32.72. Avg. time spent on learning 91.5 90.0 87.3 88.43. Achievement of learning objectives 67% 73% 60% 73%4. Discrepancy of WAVI-factors (VICS/AdeLE) 0.92 0.79 0.81 0.89

Table 8.1: Characteristics of the students’ learning behaviour for each initial WAVI-group

Table 8.1 also points out that the learning time (the time period from requesting the first instruction toreaching the last one) as well as the achievement of learning objects (percentage of students who reachedthe last instruction) did not significantly differ for the four initial WAVI-groups. On the other hand, theaverage number of clicks per student and the average discrepancy of the WAVI-factors between the AdeLEsystem and the VICS test seem to be slightly better for analyst-imagers and wholist-verbalisers.


Analysing the results of the AdeLE prototype, it has to be admitted that the simple algorithm to updatethe WAVI-factors (see section 7.3) diverges rather strongly from the results of the other two sources and,therefore, might be wrong. Further, the results of the VICS test assume that the 60 test users (studentsof a course in a technical study) are not differing enough by means of their learning style or that theWAVI-model is not applicable for adapting the learning process in general.

Concerning the utilisation of the Dynamic Background Library (see also chapter 9), another rele-vant aspect for the AdeLE project is the application of this tool. Analysing the usage of the DBL, stu-dents mainly needed translations (54 requests to LEO dictionary) and accurate information (31 requests toGoogle), while 23 requests were tracked to Answers.com, 22 to the English version of Wikipedia and 12to the German version of Wikipedia. The German version of Google (4), Google Images (2) and GoogleNews (1) were hardly used.

By means of concrete numbers, each concept defined for the course on “Adaptive E-Learning” was,on average, clicked 4.66 times, whereas adaptability (27 clicks), e-learning (19) and adaptivity (13) wererequested most frequently. Drawing conclusions from these numbers, the DBL might be of interest as atool for the provision of context-aware background information. Referring to [Modritscher et al., 2005],the DBL is applicable for different educational scenarios, which might also be a target for further adaptivebehaviour of the AdeLE system.

8.2.3 Usability aspects

The learning behaviour and also the system’s usability were strongly influenced by the overall performanceof the AdeLE prototype, which differed within the three passes as stated above. As can be concluded fromtable 8.2, first pass students did not interact so intensively as the students in the other runs, but they spentsignificantly more time on learning. Further, most of them did not complete all instructions of the system.On the other hand, students in the first and second pass clicked on the links provided by the DynamicBackground Library much more, i.e. to bridge the waiting time until the next instruction.

Characteristic Pass 1 Pass 2 Pass 31. Average number of clicks per student 27.0 28.1 34.32. Average time spent on learning 95.6 85.5 87.03. Achievement of learning objectives 15% 91% 100%4. Discrepancy of WAVI-factors (VICS/AdeLE) 0.89 0.91 0.745. Average usage of background information 2.55 3.27 1.44

Table 8.2: Characteristics of the students’ learning behaviour for each pass

Another key finding of this study deals with the users’ understanding of the adaptation process. Sincemany students stated in the post-questionnaire that they did not recognise or understand the adaptivebehaviour of the platform, providing more meaningful information about the user profile (the WAVI-factors as a basis for the system’s adaptation decisions) and learning progress (the so-called learningstates) was targeted. As a result of this study, the area “Why this way?” (see section 7.4) was added to theformer prototype.

Drawing conclusions from this case study, it can be stated that the adaptation model (the WAVI-factors)might not be ideal and still requires investigation and new studies, particularly driven by psychologists.In fact, AdeLE’s theoretical approach and the system’s architecture are extendable and flexible enoughto support further research activities. On the other hand, problems with the system’s performance couldbe solved and the prototype was improved by means of a more scrutable adaptation process. Finally,the Dynamic Background Library was identified as a powerful tool for the online learning process itself.

8.3. Other results from literature 129

Therefore, chapter 9 exclusively examines the usefulness of this tool in the context of adaptive e-learning.

8.3 Other results from literature

The study of the AdeLE prototype and, particularly, its adaptation model could identify some interestingfindings about online learning and issues like the performance of a system, a non-scrutable adaptationor the application of a Dynamic Background Library. In addition, this section outlines a few, exemplaryexperiences given by literature and practice.

8.3.1 Approaches and models for adaptive e-learning

From a theoretical viewpoint, the necessity for considering selected historical streams outlined in section4.1 is already proven, for example as also shown with the inspection of e-learning standards and systemsin the practical part of this dissertation. The content-based aspects of the macro-adaptive instructionalapproach are based on didactical requirements and, therefore, state-of-the-art specifications (e.g. IMSSimple Sequencing) and e-learning platforms. The inspection of research projects and companies in thescope of adaptive e-learning in chapter 6 pointed out that these basic concepts are realised in all prototypesand products.

On the other hand, the benefit of aptitude-treatment interaction or micro-adaptive instructions is onlyproven for selected learner characteristics or for a very restricted domain or context. Hereby, [Tobias,1989] identified several dependencies of the ATI approach, so that adapting to these variables is, in mostcases, not effective at all. Similarly, ITSes provide valuable techniques to adapt the learning process, yet[Park and Lee, 2004] state that such solutions often fail to consider important learning principles or workonly for a very restricted domain due to a lack of intelligent behaviour, for example achievable by AItechniques.

As a result, streams like Adaptive Educational Hypermedia (AEH) tried to combine intelligent tutoringwith content-based adaptivity. Such micro-adaptive instructional approaches focus on a few selected issuesof adaptation information and techniques. Furthermore, these younger research and development activitiesalso try to apply pedagogical aspects like constructivistic learning or collaboration (see also sections 4.1and 4.2). Overall, many of these approaches have been evaluated already. The results of these evaluatedapproaches will be summarised for a few selected systems in the next two subsections.

At this point, two approaches to theoretical frameworks should be addressed here: Firstly, such modelslike FORMABLE, but also any other formal or informal framework introduced in section 4.3 can be ofenormous advantage for designing and evaluating adaptive behaviour in e-learning systems. For instance,FORMABLE was utilised to build up requirements for standards and specifications as well as on systemsin order to support adaptive e-learning. Further, [Henze and Nejdl, 2004] describe various existing systemsin the field of adaptive hypermedia by applying their model for a logical characterisation of such systems.

Secondly, theoretical models in the field of pedagogy and psychology might be of interest for adaptingthe online learning process. One of these models is the knowledge space theory mentioned already insection 4.3. Although the creation of such a knowledge space for a course is still very costly, a few of thepossible application scenarios has been proven already. Amongst others, [Albert and Hockemeyer, 2002]report on applying the knowledge space theory for adapting the course structure, for adaptive assessmentof the learners’ knowledge state and for adaptive training. Next to these aspects, this theory could also beutilised to create course packages considering didactical dependencies as well as all possible paths throughthe course.


8.3.2 Techniques for adapting the learning process

Despite these streams and theoretical frameworks for adaptive e-learning, many researchers already provedthe applicability of single techniques in practice. A few results from evaluation studies are highlightedhere [DeBra et al., 1999a]:

• The adaptive stretchtext system MetaDoc by [Boyle and Encarnacion, 1994] stretches and shrinkspage fragments on the basis of the user’s preferences, i.e. derived from the click behaviour. Astudy showed that users found answers to comprehension questions significantly faster with thissystem than with traditional hypertext, while they also could give significantly better answers tothe questions. Nevertheless, no significant differences were found regarding the performance insolving search and navigation questions (search correctness, number of visited nodes and numberof operations).

• [Brusilovsky and Pesin, 1998] evaluated the adaptive link annotation and removal mechanism of theISIS-Tutor. ISIS-Tutor uses the dynamic state of user knowledge to divide task nodes into severalclasses, such as “not ready to be learned”, “ready to be learned”, “well-learnt” and “in work”.Adaptive link annotation was used to show classes of pages behind links, i.e. by adapting colourfonts and special symbols. A study showed that the overall number of navigation steps and thenumber of unforced repetitions of task pages were significantly lower with the adaptive version ofthe system (about twice as fast with only half of the steps). No difference was found between the linkannotation with or without the additional link removal technique. Another study [Brusilovsky et al.,1998] highlights further findings: Firstly, learners preferred using the continue-button (over 90%)than the annotated link. Secondly, link annotation encouraged the novices to use non-sequentiallinks more often. Thirdly, students who followed the system’s guidance were able to achieve bettertest scores.

• [Weber and Specht, 1997] examines the effect of adaptation within the ELM-ART system whichprovides adaptive guidance and annotation techniques. This informal evaluation conducted by ques-tionnaire outlines that learners worked twice as long with this system and, further, preferred one ofthe two adaptive mechanisms. While novices ignored adaptive annotation, experiences users didnot use the guidance provided and adapted by the system.

• Interbook, the authoring tool for ELM-ART, was evaluated in order to find the value of adaptiveannotation [Brusilovsky and Eklund, 1998]. Therefore, two groups had to work with the system,whereby one group experienced it with and one without adaptive annotation. Although no differ-ences between these two groups were identified, about 80% of the clicks were made with sequentialnavigation buttons, while adaptive annotation encourages non-sequential navigation. Yet, adaptiveannotation can be a benefit to those who use it as expected.

Concerning the WAVI model utilised in the AdeLE prototype, the following three evaluation studiescan be outlined:

• [John and Boucouvalas, 2002] describe an experiment that measures which cognitive style influ-ences the performance of users when listing to or viewing information presented by computers. Asa result, this case study points out that there are differences in performance when information is pre-sented visually or using audio, but significant correlations of the other dimensions (wholist-analyst,verbaliser-imager) were not observed.

• On the other hand, [Bajraktarevic et al., 2003] examined the cognitive load for wholists (globalusers) and analysts (sequential users). In this study, learners achieved significantly higher scores if

8.3. Other results from literature 131

navigational elements are adapted to this cognitive style dimension. Yet, no impact on learning timewas shown for sessions with elements matching the cognitive style and the other ones. Overall,it can be concluded that analysts might be cognitively overloaded on being displayed too manynavigational elements, such as a tree-view, links to background information or information about ascrutable adaptation process.

• Finally, [Phillips, 2005] outlines that students are bad judges of their own styles. In his study, heassessed the WAVI-factors of two different psychological tests (VVSR, CSA) and of a questionnairefor self-assessing these values. As a result, all three results did not correlate, whereby particularlythe self-assessment differs strongly from the calculated values of the two tests.

As a conclusion, the findings of the study evaluating the AdeLE prototype are strengthened and theWAVI-model seems to be inappropriate for adapting the learning process towards pedagogical issues.Nevertheless, evaluation results often addresses only selected adaptive techniques or are restricted to anarrow scope. In practice, only a few systems proved appropriate benefits for technology-based learningand teachings, as shown in the following.

8.3.3 Adaptive e-learning environments in practice

Focussing on full-featured systems for providing adaptive e-learning courses, many systems already ex-ist. Most of them can be characterised as results of research projects or long-term research activities,only a few comprise a commercial product. In the following, some adaptive e-learning environments areexamined in terms of evaluation results.

The first platform to mention is an implementation of the KnowledgeTree architecture, successors ofELM-ART and ISIS-Tutor as already introduced in section 4.3 or outlined in the last subsection. A pro-totypical demonstrator accessible at [TALER, 2002] includes various activity servers, the KnowledgeSeaportal as well as the student modelling server CUMULATE. [Brusilovsky et al., 2006] conducted a studyof the motivational effect in two adaptive hypermedia services: QuizGuide, a tool providing adaptiveannotation for self-assessment quizzes and NavEx providing adaptive guidance for the quizzes.

The following conclusions were drawn from this case study comparing an adaptive and a non-adaptiveversion of the system over two semesters: Firstly, users worked with the adaptive version of the systemtwice as much as with the non-adaptive one (261.21 vs. 127.68 activities, 23.97 vs. 13.11 quizzes, 10.18vs. 8.70 sessions). Secondly, there was a remarkable difference in using the navigational elements withand without providing adaptive guidance via NavEx (171.90 vs. 34.76 activities, 18.10 vs. 5.66 examples,8.20 vs. 3.52 sessions). Overall, this study proved that the learners work with the adaptive version ofthe system more intensively outside the so-called teaching stream (the time frame in which a teacherpresents the content and encourages learners to deal with the course’s topics). Thus, this approach seemsto strengthen the students’ intrinsic motivation and enable self-directed learning.

In the scope of commercial products only a few full-featured systems providing adaptive behaviourexist. One of them is iDL (Intelligent Distance Learning) Systems, as already introduced in section 6.4.Beside an excellent theoretical background, the website of this company also reports on various projectsand courses realised in practice [iDL, 2007b]. For instance, [iDL, 2005] reports on a significant reductionof the drop-out rate, since their solution is utilised for the IC3 Certification course. On the other hand,[Sonwalkar, 2002] points out the pedagogical necessity to adapt instruction, in this case towards the so-called “learning cube”.

Another commercial solution to examine here is ALEKS [ALEKS, 2007], a web-based, artificiallyintelligent assessment and learning system utilising the knowledge space theory. The theory behind thisproduct focuses particularly on adaptive assessment and optimising the learning path. [Falmagne et al.,2003] evaluate this approach on the basis of an extensive example comprising 397 problem types from


various areas of mathematics, such as arithmetic, introductory algebra, intermediate algebra, pre-calculus,etc. For a subset of only 88 skills (from one of these areas) it was shown that the possible number ofknowledge states can be reduced from 288=3.1 x 1026 to approximately 60,000 by applying informationabout prerequisite relations between these skills and, further, that these 60,000 possible states can beassessed by approximately 25 questions.

Concluding from this literature survey on evaluation results of theoretical frameworks, techniquesand systems, it can be stated that adaptive e-learning approaches aim to improve the learning process,whereby either the efficiency (provision of best fitting instruction for the learner’s understanding) or theeffectiveness of learning (optimising learning by means of time efficiency) is addressed.

8.4 Conclusions

To sum up this chapter, the AdeLE prototype has to be evaluated from two sides. From a technologicalviewpoint, the architecture and implementation details provide a powerful framework for realising anykind of adaptation for the online learning process. On the other hand, the usefulness of the overall proto-typical implementation is given due to the fact that online courses can be realised, which was shown bythe evaluation study depicted in this chapter.

Nevertheless, the benefit of automatically adapting the online learning process is not proven for theAdeLE system at all. The adaptation model including an adaptation towards and of the WAVI-factors wasslightly improved by this study, but still is rather poor, as no deterministic behaviour towards the cognitivestyles of the learners can be identified. Thus, more studies or another psychological approach is requiredto significantly enhance the adaptation model. Furthermore, performance issues lead to a few interestingfindings on the usefulness of the AdeLE system, but also influenced the main objective of the study, i.e.an examination of the usefulness of the AdeLE adaptation model, in a rather negative way.

However, the study conducted at the Technical University of Technology had a positive impact onseveral aspects of the AdeLE prototype finally. Firstly, the problems with the systemic performance couldbe solved completely. Secondly, the results of the study were considered by implementing new featureslike the scrutability area containing information about the adaptation model and process, the learner statemodelling cycle, an improved behaviour of the dynamic background library, etc. Thirdly, the adaptationmodel was also modified in terms of decreasing the adaptation steps of the WAVI-factors during a learningsession.

Overall, the AdeLE approach comprises a technological methodology towards adaptive e-learning,providing a powerful and flexible framework for realising other methods and techniques. As also shownthroughout this dissertation, the FORMABLE model might be of benefit for designing and evaluatingadaptive features within e-learning environments on the basis of learning content, pedagogy and didac-tics. Nevertheless, the results found in literature and practice show that adaptive behaviour of e-learningsystems has to be planned and evaluated according to pedagogical principles in order to guarantee anutilisable platform.

As a benefit of the adaptation process within the overall AdeLE prototype has not been establishedso far, the next chapter focuses on one particular and very promising part of the system, the DynamicBackground Library.

Chapter 9

Utilising a Dynamic Background Libraryfor Adaptive E-Learning

“ To err is human. Just ignore the errors and go ahead. ”

[ M. K. Soni ]

While the last chapter manifests that AdeLE’s adaptation model is rather prototypical and the appliedadaptation techniques might not really help online learning, the research project dealt with another in-teresting idea. Derived from the FORMABLE model, one of the three adaptation methods addresses theinsertion of new instructions into the existing course. Yet, this concept is not new at all and has a longhistory, as can be concluded from examples like providing static background information via links orretrieving instructional content from other sources or on-the-fly.

Instead of just displaying a static link list given by the teacher or retrieving the whole course contentfrom repositories, a flexible solution guaranteeing accurate information as well as didactical awarenesswas achieved. Therefore, the idea of a Dynamic Background Library (DBL) introduced in [Dietingeret al., 1999] and realised as a prototype named EHELP [Garcia-Barrios et al., 2004a] was evaluated forthe AdeLE approach in order to integrate this tool. Section 9.1 outlines the concept and the first realisationof the Dynamic Background Library as a basis for the evaluation study. Section 9.2 summarises a casestudy conducted at the Graz University of Technology in 2004 and section 9.3 describes the redesign ofthe DBL for the AdeLE research project.

9.1 Basic concept and realisation of EHELP

Before introducing the idea of a Dynamic Background Library or the prototypical implementation itself,the necessity of such a solution is pointed below.

9.1.1 The need for a Dynamic Background Library

As mentioned above, many systems – even from the first generation of adaptive educational hypermedia– support such features like presenting links to background information. For instance, ELM-ART andInterbook [Brusilovsky, 2000] allow teachers to define links to resources related and relevant to a course.This link list might also be adapted by techniques like adaptive link annotation or hiding, yet the links arestill given by a static link repository. Further, these links always point to the same target resources.

133

134 9. Utilising a Dynamic Background Library for Adaptive E-Learning

As a result, static links from a static link-set might lead to information which is no longer accurate.As a side-effect, such solution approaches are also vulnerable to the problematic aspects of broken links.Moreover, in reference to the well-documented rapid growth of knowledge, the selection and mainte-nance of relevant background information will be an increasingly serious challenge for both teachers andlearners.

In addition, teachers often use learning materials and background information from preferred sources,in most cases even from one particular repository. Thus, learners receive only resources from this re-stricted set of repositories. However, studying from one particular learning source will not satisfy a deepunderstanding of new knowledge and not enhance the knowledge acquisition process in terms of overalllearning efficiency.

Finally, it has to be stated that other architectural models or realised systems only partially solve theseproblems, as they provide either fully structured learning content, for example with adaptive links as de-scribed by [DeBra and Ruiter, 2001], or with totally unstructured course materials, such as the KnowledgeSea portal by [Brusilovsky, 2004b], where the teacher just specifies the range of learning materials oractivities and the system organises the content in a matrix-based knowledge map.

9.1.2 Functionality of a DBL

Due to the problems depicted in the previous subsection, the Dynamic Background Library addresses theusage of knowledge repositories for the retrieval and delivery of (internal and external) materials, whichare relevant to achieve mastery of given learning objectives and, at the same time, fit the learning context.Against this background, the concept of the Dynamic Background Library was developed at the Institutefor Information Systems and Computer Media (IICM) at the Graz University of Technology [Dietingeret al., 1998] [Garcia-Barrios et al., 2002].

Generally, a DBL improves the goal-oriented knowledge transfer process by enabling the develop-ment of a dynamically indexed background library of subject-relevant resources residing outside the staticrepository. The basic functionality scheme of a DBL, as shown in figure 9.1, depicts the different interac-tion layers and its dependencies through the knowledge transfer process:

• Beginning with the layer at the top of the figure, the section “Adaptive background knowledge ac-quisition” describes the process of dynamically retrieving resources for a pre-defined concept fromgiven repositories. Therefore, the knowledge acquisition process is implemented by means of acommunication layer to search services, such as xFIND – the abbreviation for “Extended Frame-work for INformation Discovery” [xFIND, 2003] – or Google.

• This background knowledge is abstracted within the DBL as a set of “items” (concepts), which arebasically defined for several expertise level groups and, further, assigned to one or more instructionsor even instructional fragments. The set of DBL items represent the “Semantic Knowledge Factory”,as illustrated in the middle layer of the figure. Course creators can define specific concepts andassign a query for particular search engines to each of them. For example, the teacher might wantto use Google to provide up-to-date information from the internet, the online Oxford AdvancedLearner’s Dictionary to present explanations of terms or the LEO English/German Dictionary totranslate single words or phrases.

• Depending on the selected viewing mode, the requested page is dynamically generated and containsa list of valid hyperlinked items (compare with the section “Adaptive Content Delivery” in figure9.1). Thus, clicking on a delivered item leads to a search request to the information retrieval systemusing the pre-stored and concept-specific query term. In the figure this process is symbolised by thearrow “Activation”. The final result is a set of accurate, relevant and up-to-date documents.

9.1. Basic concept and realisation of EHELP 135

Figure 9.1: Basic functionality scheme of a DBL [Garcia-Barrios et al., 2002]

9.1.3 The prototype for the evaluation study

Based on the idea of a dynamic background library, a running prototype named the Enhanced E-LearningRepository Manager (EHELP) was developed at the Institute for Information Systems and Computer Me-dia (IICM), Faculty of Computer Science at Graz University of Technology [Garcia-Barrios et al., 2004a].

Addressing adaptive content delivery from the previous subsection, the modules and instructions of acourse can be interpreted as static content. If an assigned concept is found within one instructional unit,its correspondent background knowledge, which should be accessed through hyperlinks, is dynamicallyadded to the courseware and delivered by means of the four following view modes:

• Embedded hyperlink: The content of the requested instruction is parsed and modified dynamicallydepending on the current settings, as depicted in figure 9.2 (see also bottom-left side in figure 9.1,View A). Each match is highlighted and hyperlinked to a context chooser and, further, to the IRS,which processes the corresponding search query requests that are passed through the hyperlinked


book-icon.

• End of page: A list of the matching items is appended “at the end” of the current instruction (seebottom part of figure 9.1, View B).

• End of chapter: Instructions are not modified, but at the end of each module a dynamically generatedHTML page containing an alphabetical list of the instruction and level-specific EHELP items isadded, as illustrated in figure 9.1, View C.

• End of course content: A dynamically generated HTML page with a list of all level specific itemsis attached at the end of the course (see figure 9.1, View D).

Figure 9.2: EHELP viewing mode “embedded hyperlinks” [Garcia-Barrios et al., 2002]

As stated in [Garcia-Barrios et al., 2002] and [Garcia-Barrios et al., 2004a], the EHELP prototype isrestricted to manage only one IRS, in this case Google. Further, it is realised as an extension of Hyper-wave’s eLearning Suite (eLS) [Hyperwave, 2007] and fully written in server-side JavaScript following anobject-oriented approach.

9.2 Evaluating the EHELP system

Overall, this first prototype of the Dynamic Background Library was evaluated in order to examine itsapplicability for the AdeLE prototype.

9.2.1 Setup of the study

From the technical perspective, the experimental set-up was based on the prototype implementation asdescribed in the last section (in this case the Google search system was chosen as the IRS). Participants

9.2. Evaluating the EHELP system 137

in the experiment were instructed to consume the online lessons using their own notebooks with a recentversion of the MS Internet Explorer browser.

Form the content perspective, a course on “language guessing” was created as an online lecture inthe German language. This course consists of 18 HTML pages divided into four modules. In addition,28 different search concepts (i.e. EHELP items) were prepared for the Dynamic Background Library. Acertain number of such search concepts was assigned to each of these modules: Introduction (14 items),Introduction to Linguistics (6 items), Language Guessing Methods (11 items) and Language GuessingTools (3 items).

The experimental environment as well as the experiment time slot was the same for all learners. Theworkflow of the experiment was conducted as follows (the time slice for each process step in minutesis depicted in parenthesis): introduction into the learning platform and EHELP (15), answering pre-questionnaire before online lesson (15), consumption of online lesson (120), answering post-questionnaireafter online lesson (15), short written examination to assess knowledge acquisition (15).

From the pre-questionnaire some learner attitudes and behavioural characteristics were inferred, likeinformation about computer utilisation, habits regarding general information consumption and preferredmedia while learning. Relevant information about the learner’s behaviour while learning was collectedthrough the EHELP logging features. The post-questionnaire represented a usability test concerning is-sues about personal background knowledge perception and feedback about the usefulness of EHELP. Allaspects depicted so far were combined to gain a deeper evaluative insight. The next section briefly de-scribes some of the most relevant results and findings concerning the evaluation of EHELP.

9.2.2 Statistical information about the participants

The learner group of the experiment consisted of 14 males. All of them were either bachelor or graduatestudents in the subject of software architecture or computer science. Ten learners (71%) pertained to theage class “15-25” and four (29%) to “26-45”. Almost everyone in the learner group indicated that theyuse computers extensively for work or learning activities, as well as for personal interests. Twelve learners(86%) use their computer every day at least for one hour, four of them for more than six hours a day forworking or learning activities. Again, 13 students (93%) use their computer daily at least for one hour forprivate activities, two of them for even more than six hours.

The usage of specific Internet services can be characterised as follows: The consumption of the mainservices e-mail, WWW and search is for nearly 100% of participants “repeatedly daily”. The usage ofnews fora shows a decreasing tendency. The most noticeable issue in this scope is the fact that a majority(57%) uses digital libraries merely “occasionally”, while 21% at least “daily”.

Concerning the comparison between the consumption of different media types on paper and on screen,the following statements about the learner group can be manifested here: More than 90% of the subjectsprefer to read news from screen and more than 70% prefer to consume encyclopaedias on screen. Onthe other hand, books, newspapers and lecture notes are preferably read on paper. Consequently, it canbe concluded that the learners like to use the Internet for consultation and in order to acquire currentinformation. This may be one indicator for the need and application of background knowledge.

Addressing the learning behaviour, the most remarkable issue in this context is the fact that the major-ity of the participants primarily learn on the basis of the lecture notes. The students do not really consultjournals for learning, but they tend to consult colleagues or friends and include them into their learningsessions. The common and regular usage of books has proved itself again. In addition to books and lec-ture notes for their learning activities, learners highlighted the usage of background knowledge from thefollowing sources: current online information, web sites, Internet inquiry, abstracts of lecture notes, testexamples, literature references and lecture notes from other universities.

So far, it can be stated that the participants of this study prefer to learn from lecture notes and books on


paper, but are very interested in acquiring additional (relevant and up-to-date) background knowledge anduse these additional resources within their learning activities. Thus, it seems to be promising to exploitonline resources for background knowledge supply.

9.2.3 Inferences on the usability and the usefulness of EHELP

Overall, some valuable experience for the applicability of the EHELP system and, further, for its redesignfor AdeLE could be gained. Issues about the usability and efficiency of EHELP could be extracted fromthe analysis of the post-questionnaire. Apart from the fact that the usability of the overall system and thee-learning platform was rated slightly positive with 3.8 out of 6 points, EHELP was rated with an averageof 5.1 points, whereby 43% of the students gave EHELP the highest score.

Further, positive feedback could be gained while asking for the efficiency of the information retrievalprocess through the EHELP prototype. The most important findings were identified through the “free-text”questionnaire fields in this context, however. Thus, referring to the subject usability, participants of thestudy stated the following:

• On the one hand, they pointed out that they liked cross references directly embedded in the text,the idea of graded information behind DBL items according to user’s expertise and the ease of thesystem’s usage.

• On the other hand, students also manifested that they did not like the intrusive in-text hyperlinksand the necessity of manually modifying queries in order to find more useful information.

The participants of the study gave feedback in terms of recommendations for improvements and help-ful suggestions. For instance, they pointed out that embedded hyperlinks should not be marked obtrusively,as this confuses users. Further, they mentioned that modifications of the Google index may have a negativeinfluence by means of retrieving inappropriate documents. Moreover, Google results should be displayedonly in the user’s language. A few of them even requested a direct link to cited literature resources as wellas the provision of links to online lexica like Wikipedia or the LEO dictionary, which should also be keptseparated from the Google search results.

The last part of the evaluation deals with the benefit of the DBL and the user interaction preferences.A positive tendency may be concluded from the requested personal estimation about the improvement oflearning activities through a DBL. The average rate was equal to 3.4 out of 6 points. In addition, studentsaddressed the importance for the DBL users to have control over the personal settings. More precisely,the participants wanted to select their preferences themselves and to modify the settings according to theirown will and convenience. In the special case of EHELP, users wanted explicitly to choose and changethe settings for their pre-knowledge (e.g. domain expertise) and DBL viewing modes (e.g. presentationmodes for DBL items).

Questioning the personal benefits of using the EHELP system, the participants of this study outlinedthe following advantages: Some of them could benefit from user-tailored and up-to-date tips about relevantand additional literature, while others mentioned accurate information and alternative perspectives onthe subject from different authors. Further, students pointed out the correction of individual knowledgegaps (e.g. expertise knowledge, other explanation models or foreign words), the support of autonomousinvestigation work, rapid retrieval of specific terminology and EHELP’s customisability (e.g. by selectingthe domain expertise).

Finally, referring to the overall functionality of the EHELP framework, the participants stated thefollowing: They particularly liked that explanations appear where they are needed, the possibility ofchanging personal settings and the clear structure of the system. On the other hand, students disliked thedesign of DBL links within the text, the retrieval of irrelevant documents within the Google search results

9.3. The Dynamic Background Library for the AdeLE prototype 139

and results which are partially useless or require a long further search within a website. Overall, mostof them thought that searching with Google is too global and a manual modification of the query wassometimes necessary.

For example, students recommended the enhancement of the design of the in-text DBL links, theutilisation of tool tips for links and the movement of term definitions not having influence on topicalityto a static glossary. Furthermore, the participants suggested the provision of a built-in library in orderto find topic-specific explanations, improvement of the load seed of parsed pages, presentation of wordexplanations from additional encyclopaedias (e.g. from Wikipedia) and a better structure of links, if thereare too many defined for a page (i.e. in the “End of Chapter” viewing mode).

9.3 The Dynamic Background Library for the AdeLE prototype

As the DBL was redesigned and implemented for the AdeLE system, the evaluation results of this studywere considered and lead to a more powerful version of the EHELP system.

9.3.1 Inferences for the new version of EHELP

Generally, it can be assumed that a Dynamic Background Library has a beneficial applicability for thelearning process as well as for the maintenance of learning content. Based on the evaluation of EHELP,the following inferences on the redesign of the system can be derived from the participants’ commentsafter absolving the course enhanced by the Dynamic Background Library (using Google as backgroundIRS):

• Some test users requested language-restricted results gathered by the IRS to avoid Russian or Chi-nese search results, which happens quite often with Google.

• Other test users wanted to get results from other search engines for different reasons, such as gettingdefinitions from Wikipedia, translations from the LEO online-service, or simply disliking the resultsfrom Google, because they were partially irrelevant or too global.

• Further, the majority of subjects claimed that they had to adjust the pre-defined query to find ap-propriate material, which justifies the idea of adapting the behaviour of the background library withrespect to the context. Thus, it is absolutely necessary to use the right IRS for a given context anda personalised query in order to enable the student to fulfil the postulated task. For instance, sev-eral test persons, mainly researchers or students writing on a thesis, asked for built-in links to citedliterature resources.

• In addition, some test users wanted the EHELP system to embed cross references within the courseand to other courses. Thus, some learners wanted to go through the learning content at their ownpace and in their own way, which is proven by many studies within the field of the macro-adaptiveinstructional design (see section 4.1).

These few experiences outlined here are just restricted to the view of the learners. Active lecturesapplying distance teaching methods would appreciate a tool allowing us to evaluate and update our coursematerials. Thus, EHELP is also being redesigned and extended in terms of context-dependent adaptationof the system’s behaviour, such as using different IRSes as well as features to evaluate and manage learningresources.


9.3.2 Redesign for the AdeLE prototype

From a technological viewpoint, section 7.3 already gave an overview of the AdeLE version of the DBL.In short, EHELP was ported to Java as an enhanced independent system (the Concept-Based ContextModeller in combination with the Background Knowledge Repository), which manages concept sets forfreely-definable information spaces.

The objective of this new system is to provide enhanced EHELP-features for the AdeLE prototypewithin the service-oriented approach [Gutl et al., 2004]. Yet, the implementation of the DBL was sim-plified by means of providing only one viewing mode (in the navigational area on the left side of theinstructional content) and not differentiating between novices, regulars and experts.

Furthermore, the AdeLE approach aimed at providing more general courseware, which is structuredby means of the most important high-level concepts within the adaptive e-learning environment. The newrequirements given from the evaluation study of the last section deal with structuring the course (at least byspecifying instructions), assigning concepts to instructions and defining queries for different informationretrieval systems.

Figure 9.3: Background knowledge data structure

Figure 9.3 describes the data structure of the background knowledge to be generated for a course.Thus, a teacher can create the context (the course) containing a list of so-called context-items (the instruc-tions). Further, all concepts as well as a list of IRS with their specific query terms have to be defined.Thereafter, the teacher can assign relevant concepts to each instruction. Each concept is automaticallyassociated with each IRS, whereby the teacher can restrict the information sources for each concept witha white list.

The following subsection outlines how useful such a system could be within the scope of adaptivee-learning, by pointing out a set of possible application scenarios.

9.3.3 Opportunities for adaptive e-learning

Based on a literature research, several possible scenarios for the application of a Dynamic BackgroundLibrary in the field of technology-based learning or any educational activity can be identified. Six of thesepedagogical scenarios are listed and explained as follows:

• First of all, [Campbell et al., 2004] reports on students from abroad having problems with under-standing the language, in particular certain vocabulary. In this context, a DBL could offer useful

9.3. The Dynamic Background Library for the AdeLE prototype 141

translations applying some kind of dictionary service. With respect to the research project AdeLE,the eye-tracking device could be applied to detect if a learner from a foreign country – in terms ofnot speaking the course language fluently – has problems reading or comprehending a word or aterm. By exploiting the eye-tracker characteristic of also being a gaze-tracker, linguistic problemsmight be identified, for example as a result of the scanning path, a high frequency of returns tospecific fixation places or the reading velocity.

• Similarly to the previously illustrated scenario, a learner could have problems understanding a pas-sage in general due to a lack of knowledge about a certain concept (word or phrase), unclear andcontradicting formulations or, again, completely unknown terms. Using EHELP to retrieve relevantand context-specific sources for the problematic passage (for instance by querying a digital repos-itory as described in [IMS, 2003]), the learning process could be improved by offering the studentanother explanation model or some missing definitions. At this point it has to be stated that in or-der to prevent an unmanageable overload of information the teacher has to pre-define some kind ofcriteria for the relevance of the material.

• Another aspect which is addressed [Dreher et al., 2004b] regards thematic-driven learning, com-prising the idea that each student specialises in one chosen topic and then reflects other topics byperforming certain tasks such as peer reviewing or peer assessment. Thus, EHELP would be a greattool to manage and provide the topics as well as different materials for the students to study in amore focused way. Again, a repository for learning objects, as one of the IRSes in the background,would be of high relevance.

• Referring to the constructivistic theory outlined in section 3.1, new paradigms in the e-learningsituation are of importance nowadays. Context-driven learning can be supported by EHELP throughthe usage of different IRSes to retrieve background knowledge. For instance, one context is reflectedby the scenario already mentioned, where the classical learning process of a student takes placeand EHELP offers background knowledge, such as definitions or alternative explanation models.Another context could comprise a researcher scanning the learning content for new developmentsthrough applying a specialised IRS on the internet in order to gather accurate information. A moreneutral context could be the evaluation of content through a teacher, where no certain backgroundknowledge needs to be retrieved.

• Adaptability and even adaptivity addressed in section 2.3 may be realised by defining differentconcepts for different types of learners or for learner groups, based on user information such as pre-knowledge in the domain, preferred search engines, or interests. Furthermore, it is also imaginableto ignore the course structure and pre-defined sequencing completely and allow learners to explorethe course freely and at their own pace. Therefore, a DBL could provide important didacticalanchors using the concepts defined by the teacher.

• Quite often, learning content needs to be updated continuously. In this regard, EHELP may alsobe applicable for teachers in order to gather new information about defined concepts and linkinginstructions to external sources or other courses. Therefore, a DBL could be used to support theevaluation of accuracy and topicality of course content by linking concepts to new information viainternet. Thus, the EHELP system has to be extended in order to interact with a learning contentmanagement system (LCMS). Again, the service-based approach to the AdeLE prototype seems tobe very advantageous.

As can be concluded, a Dynamic Background Library can be a very powerful tool within an adaptivee-learning environment. The implementation of the AdeLE system described in chapter 7 already containsa DBL, by means of the Concept-Based Context Modeller and the Background Knowledge Repository.


A few of the pedagogical scenarios depicted above are already supported within this prototype (e.g. thefirst two), while others might be addressed by future work. The most outstanding opportunity of the DBLsolution approach comprises a more didactical approach, namely the knowledge space theory alreadymentioned in various chapters of this work.

Extending the data structure of the conceptual space by enabling relations amongst concepts wouldenable the possibility of defining concepts about skill actions (e.g. on the basis of the Bloom taxonomy)and connect these actions with course concepts. Thus, the data model would support the creation ofskill models, whereby single skills could be connected to any kind of instructions (passive information aswell as tasks). Furthermore, such a skill model could be exploited for different application scenarios likecreating or adapting the instructional sequence, adaptive assessment, etc. Nevertheless, these theoreticalconsiderations are part of future work as such an approach might be comprehensive enough to warrantanother dissertation.

9.4 Conclusions

This chapter outlined the overall idea of a Dynamic Background Library and its applicability to technology-based learning and adaptive e-learning. Inspecting approaches presenting static link-lists and other retrieval-based techniques, a DBL eliminates the technical inconveniences of data storage and provides accurateinformation by dealing with queries to information retrieval systems instead of a links.

The EHELP system, as a first prototype of a DBL, allows teachers to define concepts and a queryto a certain IRS and assign these concepts to the course’s instruction for different pre-knowledge levels(novices, regulars, experts). The adaptive content delivery unit generates instructions and includes thebackground information in different viewing modes (embedded hyperlinks, end of page, end of chapter,end of content).

The evaluation study of the first EHELP prototype showed that this first version of a DBL still has afew usability weaknesses and lacks some functions. The main critique concerns the quality of the searchresults by Google, the restriction to only one IRS as well as the viewing mode with in-text hyperlinks.Further, some interesting recommendations towards pedagogical application scenarios were given.

As a result, the new version of the dynamic background library, which was implemented for the AdeLEsystem, supports different IRSes, but provides only one viewing mode presenting the links to the back-ground library beside the instructional content. Nevertheless, the Concept-Based Context Modeller whichis implemented as one Openwings service offers new possibilities, for example for adapting the learningprocess along certain pedagogical scenarios or for managing a skill model described by the knowledgespace theory.

Nevertheless, this chapter proved that a Dynamic Background Library can be a powerful tool, appli-cable to technology-based learning and teaching. Further, the pedagogical scenarios derived and partiallyrealised within the AdeLE system showed that a DBL is advantageous for adaptive e-learning. Finally,teachers might benefit from this solution approach in terms of maintaining learning materials for coursesand specifying the underlying skills. The second aspect would, again, be beneficial for adapting the learn-ing process.

Chapter 10

The Impact of a Didactical Strategy onLearning

“ There is nothing wrong with change,if it is in the right direction. ”

[ Winston Churchill ]

The last two chapters pointed out that adaptation of the learning process might not provide benefitsfor knowledge transfer if the adaptation model lacks pedagogical or didactical validity, for example (likethe one of the AdeLE prototype), but can enhance learning in different educational scenarios, for exampleby utilising tools like a Dynamic Background Library. Consequently, this chapter examines, if and howteacher-triggered adaptation of online learning influences the knowledge transfer.

Therefore, a case study was conducted at the University of Applied Sciences Campus02 [Campus02,2007]. Precisely, three online courses were implemented with the intention to follow the ideas of well-known learning theories. After describing the planning of the e-learning study in section 10.1, section10.2 compares the three courses to each other. The courses are then examined from a didactical andpedagogical viewpoint in section 10.3.

The three different didactical strategies for realising online courses comprise a behaviouristic, a cog-nitive and a constructivistic approach. For these different e-learning methods, the following assumptionsabout didactical and pedagogical aspects were made by the author:

1. The online courses implementing these three didactical strategies may vary in the preparation, im-plementation and concluding stage for both the teacher and the learners.

2. The behaviouristic and the constructivistic approach may not be very effective and popular due tothe disadvantages of these two learning theories.

3. Group tasks may be more effective and popular than tasks for individuals.

4. The three courses differ with respect to didactical and pedagogical aspects.

10.1 Realisation of the courses regarding the learning theories

The following study was accomplished within the scope of an e-learning project at Campus02 [Sindler,2005] and dealt with an online course on the topic of “document formats”. Although the instructional

143

144 10. The Impact of a Didactical Strategy on Learning

unit can be considered as a lecture on the basics of information technology, attempts were made towardsreaching the whole range of competencies and some higher-level objectives to cover and examine a broadrange of didactical aspects.

Characterising the course with reference to the Bloom taxonomy, the educational objectives mainlycomprised imparting knowledge on the students, but also included two skills and one affective goal, asshown in table 10.1. When planning this study, the lecture was implemented in three different onlinecourses, each one realising the didactical strategy related to one learning theory described in section 3.1.Therefore, the 38 students were split up into three groups according to the students’ performance in aprevious lecture related to the topic and assigned to the courses.

Domains according to Bloom Taxonomy Level 1 Level 2 Level 3Cognitive Domain 5 4 2Psychomotor Domain 0 0 2Affective Domain 0 0 1

Table 10.1: Statistics of the course’s educational objectives

Subsequently, a customised version of the open source platform Moodle [Moodle.org, 2007] was usedto launch the courses which, then, were successfully running over a two month period. The online coursesdealt with the same learning content and, in addition, tried to achieve the same objectives depicted above.Hence, each course applied different didactical activities, as shown in the following subsections.

10.1.1 Course A, the Behaviouristic approach

Course A was planned with respect to Behaviourism, whereby learning objectives and materials wereportioned into three modules by the teacher and each of the 14 students had to study each module andfinish it within a certain period of time. The sequence of the instructional portions as well as the schedulewas given by the teacher. The students’ achievement levels were measured with an online examination.

Furthermore, this course included some playful activities, such as the possibility of several attempts atthe exam, an increasing difficulty level on later modules, one task to gain a bonus, etc., to keep the learnersmotivated. The learning process was assessed by typical behaviouristic elements like multiple-choicequestions, assignment tasks or short answers. To examine the high-level objectives of the psychomotorand affective domain, ITS methods were simulated by the teacher, for example by manually evaluatingsubmitted strings encoded by Huffman or an LZW compression.

10.1.2 Course B, the Cognitive approach

Course B attended by twelve students was implemented according to the ideas of Cognitivism. Its taskscan therefore be characterised by classical cognitive elements, such as repeating learning content in dif-ferent ways, working out parts of the course within group work or re-structuring the content. Differentlearning styles were covered by providing different kinds of instructional support using various learningactivities of Moodle. Motivational aspects were realised by fast responsiveness of the teacher and by abonus system. Further, meta-cognitive skills of students were reflected or even enhanced by forcing stu-dents to work in groups. Finally, the students’ tasks were focussed on including own experience withintheir work.

Overall, this course was divided into two phases: Firstly, three groups consisting of four students eachhad to work out a part of the course’s objectives. In the second phase, the groups were reassembled intofour groups with three members, while each group had to restructure the results of the first phase using

10.2. Comparison of the three e-learning strategies 145

a WIKI environment. To motivate the groups, the best work of the second phase was awarded with abonus. To assess the learning process, the results of each phase were graded by the teacher based on thequality and quantity of the students’ work within the group, which is reproducible by the version-controlfunctionality of the WIKI module.

10.1.3 Course C, the Constructivistic approach

Course C comprises the idea of constructivism enforcing each of the twelve participating students to workactively on the tasks within a group of three members. Further, all kinds of interactive elements such aschat, discussion group, tasks, etc. were provided within the Moodle system and the students were alsoallowed to collaborate outside the e-learning platform. Thus, students held full control over the learningprocess and were able to manage the schedule on their own.

The approach was realised by giving the four groups all materials and the task to create a document formediating the course’s learning objectives to their colleagues. In the second phase, the three members ofeach group had to compare the works of the other groups, evaluate them by distributing a certain amountof points and explain this distribution. Again, the group with the best work received a bonus. The groupassignment work was graded by the teacher on the basis of the students’ peer reviews.

While the e-learning phase was in progress, students in the courses were instructed to document certainaspects, such as the effort for learning, a self-assessment on reaching the objectives, etc. Furthermore, anunannounced and challenging examination as well as a post-questionnaire was carried out in the courseof the lecture held after the e-learning experiment. Based on the whole amount of data retrieved fromthis study, the next section summarises the experiences gained about the different e-learning strategiescomparing them to each other.

10.2 Comparison of the three e-learning strategies

To evaluate the different e-learning methods, the three courses are compared to each other with respect toaspects such as the effort for the teacher or the students, the effectiveness and so forth. In the followingsubsections each stage of the study is examined more closely.

10.2.1 Preparation stage

First of all, the preparation stage for the study took the lecturer approximately 15.5 hours distributed asshown in table 10.2. As materials existed already, there was not much effort required to prepare the onlinematerials, which were the same in each course.


Teacher’s activities and effort (∗ all courses alike) A B C1. Determining organisational parameters 1∗ 1∗ 1∗

2. Defining the learning objectives 1∗ 1∗ 1∗

3. Preparing the existing materials 2∗ 2∗ 2∗

4. Assigning students to the three courses 0.5∗ 0.5∗ 0.5∗

5. Creating instruction for the ongoing evaluation 0.5∗ 0.5∗ 0.5∗

6. Creating instructions and activities for the course 5.5 1.5 1.57. Preparing concluding tests and post-questionnaire 2∗ 2∗ 2∗

Teacher’s overall effort [in hours] 12.5 8.5 8.5

Table 10.2: Characteristics of the three courses for the preparation stage

Including the effort of seven hours for course-independent activities, the teacher spent more time tocreate course A than to prepare the courses B or C. These differences can mainly be explained by the higheffort for creating questions with the Moodle system. Yet, the quizzes created enabled rapid grading andcan be reused for other online courses on this topic.

10.2.2 Running the online courses

Secondly, the implementation of the three courses required an amount of 11.5 hours of work from theteacher. An overview of the teacher’s activities and efforts for carrying out each course is given in table10.3. In addition, students in course A thought they were able to master most of the 14 learning objectives,while students in course C slightly doubted it and students in course B were very pessimistic about theachievement of the defined competencies. Moreover, student groups in courses A and C decided to workon the tasks individually most of the time, while the participants in course B opted to work with otherstudents at least 30% of the time.

Teacher’s activities and effort (∗ all courses alike) A B C1. Introducing the online course in the lecture 1∗ 1∗ 1∗

2. Weekly mail to inform and motivate students 2 1 13. Supervising the group tasks - 1.5 0.54. Individual feedback on students and group tasks 2 1 0.5Teacher’s overall effort [in hours] 5 4.5 3Further characteristicsStudents’ self-assessment of effort [in hours] 12.2 9.4 7.6Students’ self-assessment of mastering objectives 92.9% 46.8% 74.3%Students’ self-assessment of learning alone 96.9% 69.2% 98.8%Number of teacher’s activities 3278 3712 1773Number of students’ activities 2969 8037 3162

Table 10.3: Characteristics of the three courses for the implementation stage


An interesting aspect in this stage is the distribution of online activities of the teacher and the studentsin the e-learning platform Moodle. Although the online activities are only a part of the teaching andlearning process itself, the distribution of the clicks over the period of the e-learning phase might givesome interesting interpretations on different aspects of the course.

Analysing the number of activities in the three courses, course A is characterised by the fact that theteacher had more activities than the 14 students altogether (3278 vs. 2969 activities). Interpreting figure10.1, it is obvious that the number of activities amounts to up to 500 activities right before each moduleended, because the students had to pass the examination by these deadlines. As shown in the chart, thesepeaks can be found before the deadlines (21st April, 19th May and 2nd June 2005), while the activities atthe other events, as for example the intensive and course-independent discussion at the beginning of thee-learning phase, are significantly lower.

Figure 10.1: Distribution of teacher’s (blue), students’ (pink) and overall (yellow) activities course A


Contrary to course A, the twelve students in course B had to act more than twice as much as the teacher(3712 vs. 8037 activities) due to the usage of the WIKI module. Furthermore, the deadlines for the twophases (6th May and 2nd June 2005) are not clearly recognisable. The peaks of activity numbers (up to1000) can be identified in the second phase as shown in figure 10.2. This could be seen as an indicatorthat the first phase was too long, or the workload in the second phase was too high. Nevertheless, in thiscourse each student had to work with the Moodle system, which was guaranteed by the tasks and theversion-control feature of the WIKI environment.

Figure 10.2: Distribution of teacher’s (blue), students’ (pink) and overall (yellow) activities course B


In course C, the teacher’s activities were about half of the ones carried out by the twelve students(1773 vs. 3162 activities). Looking at the distribution of the activities in this course (see figure 10.3), thepeaks of the activities are distributed over the whole course and the deadlines (13th May and 2nd June2005) cannot be clearly identified. Other events such as an extensive usage of the discussion group orcourse announcements caused more activity in the course than the deadlines. Furthermore, in this coursethe possibility was offered that only one of the three group members posted the results of the first phase,while the other work was performed offline.

Figure 10.3: Distribution of teacher’s (blue), students’ (pink) and overall (yellow) activities course C

Summarising the distribution of activities over the period of the e-learning phase, the following state-ments can be made:

• The behaviouristic approach is characterised by a high degree of online learning and teaching aswell as clearly recognisable deadlines, as it is obvious that most students prefer to finish a moduleas late as possible.

• The cognitive approach (including the usage of the WIKI module to enforce students to work online)seems to cause a lot of work load for the students as well as greater effort for the teacher.

• In contrast, the constructivistic approach is characterised by a smaller effort for both the studentsand the teacher. The activities seem to be distributed equally over the whole period.

• Generally, group work can be characterised by a better distributions of activities. Besides, unbal-anced workload can be identified much better if students work in teams.

10.2.3 Concluding the study

Finally, the concluding stage of the e-learning phase took the teacher about 5.5 hours, which is distributedas shown in table 10.4. The results of the ongoing assessment were rather equal in each course due to verygenerous grading to keep the students motivated. In fact, the results of the unannounced and demanding


final exam carried out in the classroom are more reliable and allow the evaluation of the effectivenessof each course, which is strongly related to the students’ self-assessment of being able to master theobjectives (table 10.3).

Teacher’s activities and effort (∗ all courses alike) A B C1. Assessment of learning process and grading 0.5 2 12. Concluding exam and post-questionnaire 2∗ 2∗ 2∗

Teacher’s overall effort [in hours] 2.5 4 3Further characteristicsResults of the running courses’ assessment 78.1% 78.9% 79.9%Results of the concluding exam 54.8% 37.4% 43.2%

Table 10.4: Characteristics of the three courses for the concluding stage

Considering the effort for each course, it has to be stated that the most time-consuming course forboth the teacher and the students was course A. The teacher had much more preparation effort, while thestudents had to invest a lot of time due to the fact that they had to master the course on their own. Onlythe teacher’s grading was fast and easy in this course due to the usage of Moodle’s quizzes module.

Course B demanded a smaller effort from both the teacher and students, even though the grading wasmore complex. Since the students in this course spent a great amount of time on the tasks, it was obviouslynot very effective to use the WIKI module for extensive group work – the students considered themselvesto be more concentrated on the tool than on the learning content. Course C is characterised by the lowesteffort for all aspects except the grading of the group work. Contrary to course B, the peer review tasksupported the teacher in grading.

Drawing conclusions from the students’ self-assessment of their effort (table 10.3), their online ac-tivities (figure 10.1, figure 10.2 and figure 10.3) and the results of the concluding exam (table 10.4), thestudents’ workload concerning the course’s topic seemed to be at a high level in course A and at a mediumlevel in course C. Further, the students had to spend a lot of time on the tasks and had to use the platformmore intensively in course B, but the achievement levels as well as the self-assessment of mastering thelearning objectives was relatively low. As a consequence, the overall educational strategy of this courseproved to be inefficient. Although the workload of the group tasks was certainly high, the students focusedtoo much on system usage than on learning content.

Summarising the questionnaire students had to fill out, course A was rated neutral, but both positiveremarks like “a good extension for a course” as well as negative statements such as “missing explanationsfor more complex content” or “disappointment about the online course” can be found. On the contrary,course B was pounced due to the usage of the WIKI module. Students in course C were neutral aboutthe e-learning phase, but gave a few negative remarks, such as “doubt on the didactical model”, “hardeffort” or “tasks too low-level”. It has to be said that this course was less time-consuming for learnersthan the other two. The learning materials were rated as neutral, while the e-learning platform was largelyaccepted by the students. In particular, the usability of the system’s features (except the WIKI module)was highlighted as good.

10.3 Findings on didactical and pedagogical aspects

Beside this comparison of the three courses, this section deals with three further issues of evaluating the e-learning study. Firstly, the students’ achievement of the course objectives is addressed. Secondly, selectivelearner characteristics are examined in two ways, for all participants and for the students of each course.

10.3. Findings on didactical and pedagogical aspects 151

And thirdly, the factors relevant to the learning process are analysed for each e-learning strategy. For thesethree approaches, all data collected during and after the e-learning phase is exploited.

10.3.1 Achievement of the course objectives

Analysing the answers of the concluding exam (see table 10.5), it can be stated that most of the studentsattempted the questions about the objectives of the cognitive domain and especially about the low-levelobjectives, while only a few students tried to answer the question about the objective of the affectivedomain and, further, very few of them answered it correctly. Hence, this does not really guarantee that thestudents really adopted the defined behaviour.

Objective (Type/Level) Attempted (Overall, Mastered (Overall,Course A/B/C) Course A/B/C)

1. Overview of scientific working (K/1) 64.9% (64%/82%/50%) 50.9% (58%/59%/36%)2. Valuing given citation rules (A/3) 24.3% (29%/18%/25%) 9.0% ( 9%/10%/ 9%)3. Comparing layout and structure- 37.8% (29%/36%/50%) 23.3% (19%/15%/36%)oriented formats (K/2)4. Overview of text-oriented formats (K/1) 94.6% (93%/91%/100%) 73.2% (81%/65%/72%)5. Reasoning facts of text-oriented 91.9% (86%/100%/92%) 83.5% (81%/86%/84%)formats (K/3)6. Explaining colour models (K/2) 32.4% (36%/18%/42%) 17.1% (30%/ 5%/13%)7. Overview of halftone images (K/1) 83.8% (100%/64%/83%) 60.0% (74%/45%/57%)8. Explaining compression algorithms (K/2) 78.4% (86%/82%/67%) 42.1% (58%/34%/31%)9. Applying compression algorithms (S/3) 21.6% (43%/ 9%/ 8%) 9.2% (19%/ 3%/ 3%)10. Comparing graphical formats (K/2) 89.2% (93%/82%/92%) 60.8% (73%/42%/64%)11. Overview of digital audio (K/1) 89.2% (93%/82%/92%) 60.6% (65%/48%/66%)12. Overview of digital video (K/1) 81.1% (93%/73%/75%) 57.2% (79%/39%/49%)13. Designing an information system 78.4% (86%/73%/75%) 58.3% (73%/52%/48%)for different document formats (S/3)14. Reasoning the application of document 70.3% (64%/73%/75%) 37.0% (48%/22%/38%)formats in information systems (K/3)

Table 10.5: Objectives, competencies according to the Bloom taxonomy (Type: knowledge, skill, orattitude; Level: 1-6) and the rates of attempts and successful achievements (overall and for eachcourse)

Similarly, the questions on the two objectives of the psychomotor domain were avoided by most of thestudents. In particular, eight out of 38 students tried to apply the compression – six from course A, onefrom course B and one from course C. Although this question was much easier than the ones in the virtuallearning phase, none of these eight students could answer it absolutely correctly.

By interpreting the results of the three online courses, it is important to define learning objectives andto compel the students to deal with each objective. The more a student is forced to treat each objectiveon his own the better the self-assessment of being able to master this objective as well as the performancein an exam. In particular, course A worked very well because each student had to work alone on eachobjective.

Nevertheless, course C showed excellent results despite the risk that constructivistic learning mightnot work for all students. Thus, it has to be mentioned that the second task, the peer review of the othergroup works, was mainly responsible that each student treated each objective at least twice. Course B


performed very weakly as a result of a severe didactical problem. It was possible for a student to leave outobjectives within the two phases of the group tasks.

10.3.2 Assessment of the learner characteristics

The following findings on the learner characteristics are mainly taken from the post-questionnaire, wherethe students had to evaluate and rate certain statements. As literature manifests that the students’ self-assessment of learning behaviour is often wrong and psychological tests are more reliable, it has to be saidthat the statements in this questionnaire are easy to understand and formulated in the way that the studentscan hardly infer one of the learner characteristics. Furthermore, the questionnaire was performed rightafter the end of the e-learning phase, so that significant differences between the courses might be found.

Self-assessment of learner characteristics Overall Course A Course B Course C(x/s) (x/s) (x/s) (x/s)

1. Extensive prior knowledge about topic 3.7/1.0 3.7/0.9 3.6/1.3 3.8/0.92. Interest in the course’s topic 3.1/1.2 2.8/1.3 3.2/0.9 3.3/1.23a. Preferring online learning to reading printouts 1.8/0.9 1.8/1.0 1.9/1.1 1.7/0.93b. Preferring online tests to written exams 2.9/1.2 2.8/1.4 2.5/1.2 3.3/1.13c. Preferring open-ended to closed questions 3.5/1.1 3.5/1.3 3.2/1.1 3.8/0.93d. Preferring learning within a group 3.1/1.1 2.8/1.1 3.5/1.0 3.1/1.03e. Preferring interactive elements 3.0/1.1 2.8/1.2 3.2/0.8 2.9/1.34a. Acquiring knowledge by typical cognitive 4.0/1.0 4.2/0.7 3.8/1.3 4.0/1.0processes (structuring, summarising)4b. Intensively studying complex content 4.2/0.9 4.5/0.7 4.1/1.0 4.0/1.14c. Need to practice new skills 3.9/0.9 4.2/0.9 3.9/0.6 3.7/1.04d. More wholist than analyser 3.5/1.2 3.7/1.3 4.1/0.9 2.8/1.24e. More imager than verbaliser 3.5/1.1 3.7/1.0 3.3/0.8 3.5/1.05a. Understanding background by research 2.9/1.0 3.3/1.0 2.7/0.9 2.6/1.1or questions (related to Kolb’s “converger”)5b. Meaning-oriented learning of important 4.4/0.7 4.5/0.8 4.6/0.5 4.2/0.7concepts (related to Kolb’s “assimilator”)5c. Tying up to own experiences (related 3.9/0.9 4.2/0.7 3.7/0.9 3.8/1.1to Kolb’s “diverger”)5d. Finding practical examples of theoretical 4.1/0.8 4.4/0.7 4.2/0.6 3.7/1.0content (related to Kolb’s “accommodator”)6. Easy to be motivated by game-based elements 3.3/1.3 3.5/1.3 2.9/1.3 3.6/1.2or competitions7. Preferring autonomous to teacher-driven learning 3.1/1.1 3.2/1.2 3.3/0.9 2.7/1.2

Table 10.6: Results of the post-questionnaires on the learner characteristics (each statement rated witha number between one and five comprising the range from “absolute disagreement” to “strongagreement”)

The results of the post-questionnaire on learner characteristics are summarised in table 10.6. Consid-ering prior knowledge (1) about the course’s topic, most students agreed with this statement and thoughtthey had good experience and knowledge within the course’s domain. This can be reasoned due to thestudents’ experience from their jobs as well as from former lectures dealing with a few parts of the learn-ing content. The even distribution across the three courses may be a result of the manual assignment ofstudents to the courses. While the overall interest in the online course (2) is rather average, students in the


behaviourist e-learning strategy seem to be less interested in the topic than the other students that masteredthe course with group tasks.

The students’ preferences (3a-e) show some surprises. First of all, students through all coursesstrongly dislike learning on the screen. Secondly, their attitudes towards preferring online tests to writ-ten exams differ from disagreement in course B to a slight agreement in course C. Similarly, students incourse C agree more strongly on preferring open-ended to closed questions than students in course B.Furthermore, students in the behaviourist approach are slightly negative about learning in a group, whilethe students in the other courses are positive about it. Applying interactive elements seems to be neutraland evenly distributed amongst the students in all courses.

Summarising the findings on cognitive styles (4a-e), the students’ ratings for the statements wererather high, which means they agreed about acquiring knowledge by typical processes or the need topractice new skills. The self-assessment of the so-called WAVI-factors outlines that students considerthemselves to be more wholist and visualiser. The wholist factor shows strong differences between coursesA and B and course C. Nevertheless, [Phillips, 2005] states that learners are – for different reasons – notgood judges of their style and even psychological tests are not fully reliable in assessing cognitive styles.

Referring to Kolb’s learning style inventory [Kolb, 1984], four questions (5a-d) focussed on evalu-ating the students’ learning behaviour by means of being more active or reflective in the learning processas well as thinking in a more abstract or pragmatic way. Overall, the students participating in the threecourses see themselves as rather reflective learners, while there is no clear tendency for abstract or prag-matic thinking. Analysing the groups of the three courses, students in course A seem to rate their attitudestowards diverging and converging higher than the participants in the other courses. On the contrary,students in course C do not consider themselves to being good accommodators as in the other courses.Generally, these results of the students’ self-assessment are unreliable due to the reasons stated above forthe ratings of the WAVI-factors.

Concluding the results of the post-questionnaire, both game-based elements (6) as well as autonomouslearning (7) are rated neutral by the overall class. Hence, students in courses A and C are more convincedof the fact that game-based elements improve their motivational states slightly. On the other side, self-directed learning is seen much more positively in courses A and B – the two courses mainly driven by theteacher – than in course C, which implemented a constructivistic approach.

Finally, it has to be stated that background knowledge was considered by the teacher in three ways:Firstly, the course was given in German, the students’ native language. Secondly, the teacher introducedthe students to the Moodle platform in the lecture, before the e-learning phase was started. Thirdly, thestudents participated in a technology-focused study for two and a half years. Overall, the students shouldnot have had problems with the course’s language or with the usage of the system. Analysing the questionsabout the system’s usability and the quality of the learning material, these two aspects were – excludingsome problems about using the WIKI module in course B – not mentioned at all by the participants. Onthe contrary, students stated that the Moodle platform can be used easily and intuitively.

10.3.3 Analysing the factors of learning

Considering the factors influencing the learning process (see section 3.2), two sources are of relevance forthe findings of this analysis: (a) the distribution of the students’ activities and (b) the students’ ratings ofthe statements of the post-questionnaire.


Characteristics Course A Course B Course CStudents’ self-assessment of average effort [in hours] 12.2 9.4 7.6Students’ self-assessment of mastering objectives 92.9% 46.8% 74.3%Students’ self-assessment of the necessity for lecturers notes 68.8% 31.3% 62.1%Students’ self-assessment of using external material 14.3% 43.3% 29%Students’ self-assessment of learning alone than in a group 96.9% 69.2% 98.8%Results of the concluding exam 54.8% 37.4% 43.2%Overall number of the students’ activities 2696 8037 3162

Table 10.7: Characteristics of the three courses based on the students’ ongoing documentation aboutlearning and raw database queries within the Moodle system

Analysing the students’ attention within the online courses, the distribution of the course activitiesand certain characteristics is useful, but does not cover all findings requiring attention. It is not possible toinfer something about the students’ behaviour while learning offline. Nevertheless, table 10.7 (the courses’characteristics) as well as figure 10.4 (comparison of students’ activities of the three courses) documentthat students in course B had to work much more with the Moodle system than the participants of the othercourses.

Figure 10.4: Comparison of the students’ activities for the courses A (green), B (yellow) and C (blue)

Interpreting the results of the concluding examination in this context, the high degree of learningwithin the Moodle platform does not really imply a high degree of attention, because students in courseB performed worse than the others. Further, several students in this course stated that the WIKI modulelacks good usability and they had to concentrate more on the tool than on the learning content. Thus,good usability of a learning management system is an important factor for a high degree of the students’attention and adequate results in learning.

Emotional states of the students can be summarised by three observations: Firstly, the instruction toupload a photo of each student as a profile avatar caused a lot of activities in the discussion groups of all


three courses (see lines in figure 10.4 around the 5th of April). Students considered this to invade theirprivacy; some started emotionally driven discussion threads, while others refused to upload a valid photo.

Secondly, the evaluation of course B was worse than the one of the other courses due to the usabilityproblems and the higher resulting effort, as already described. Thus, several rather emotional commentssuch as “I hate e-learning”, “The task is senseless”, etc. were written down in the post-questionnaire,while students in the other two courses were not that aggressive or negative. Further, the higher effortcompared with the other courses was criticised (see table 10.7). Thirdly, the curiosity about the learningcontent was rather neutral and evenly distributed over the three courses (see statement 1 in table 10.8).

The students’ motivation for learning is examined by the following three findings:

• Firstly, in course A the deadlines (see the green line in figure 10.4 around the 21st April, the 19th

May and the 2nd June) are more obvious than in the courses implementing group works (see theyellow line around the 6th May and the 2nd June as well as the blue line around the 13th May andthe 2nd June). Due to the increasing activities ahead of the deadlines, it seems that a behaviouristicapproach enforces students to accomplish the examination right before the module ends.

• Secondly, each course included some motivational elements as stated in section 3.2 as well as somestatements about the necessity to reach the objectives. Yet, students in the courses A and C meantto master the objectives better than the participants in course B, which highly correlates with theresults of the concluding exam (see table 10.7).

• Thirdly, the motivation to complete the tasks was, according to the questionnaire’s statement 2 (seetable 10.8), average in course A and slightly negative in the courses B and C.

Self-assessment of learner characteristics Overall Course A Course B Course C(x/s) (x/s) (x/s) (x/s)

1. Curiosity about learning content 3.1/1.1 3.2/1.3 3.0/1.2 3.2/1.02. Strong motivation to accomplish the tasks 2.6/1.0 2.9/1.0 2.5/1.0 2.4/1.1

Table 10.8: Results of the post-questionnaire concerning the factors relevant to learning (each statementrated with a number between one and five comprising the range from “absolute disagreement” to“strong agreement”)

Aspects of prior knowledge were already examined in the last subsection. Moreover, the teacher con-sidered the tying up to prior knowledge by giving practical and well-known examples of theoreticallearning content as well as cross-references to other lecturers. According to Cognitivism, issues concern-ing remembering and forgetting were realised only in course B. Yet, these considerations seemed tohave failed due to the bad learning results of this course.

Finally, feedback and tutoring for each course differed depending on the three learning theories:While it was necessary to actively stimulate the participants’ learning in course A, students in course Brequired immediate feedback about the completed tasks. In course C the teacher had to suppress anycomment on the work submitted. Overall, the three courses, as well as the learning outcomes, differed inseveral aspects of the learning process, as already shown in this section.


10.4 Conclusions

To sum up this study, the five assumptions made at the introduction of the chapter can be commented onin the following way:

1. Each of the three courses varied in several aspects (such as the effort, the effectiveness, the teachingand learning behaviours, the acceptance and so forth) within the stages of preparation, implementa-tion and conclusion for both the teacher and the students.

2. The behaviouristic and the constructivistic approach showed better results for effectiveness in teach-ing and earned a better rating from the students.

3. This study showed that the best efficiency of knowledge transfer measured by the students’ achieve-ment can be obtained through those tasks that students have to complete on their own. It has to benoted that, although the first phase of the constructivistic approach was intended to be a group task,most members of the groups decided to work separately on parts of the task and merge their resultsafterwards.

4. The three courses also differed with respect to didactical and pedagogical aspects, as certain typesand levels of learning objectives were achieved by the students more successfully than others andcertain pedagogical factors, like attentional or motivational states, depended on the e-learning strat-egy. On the other hand, the didactical model of the courses had no significant impact on factors likeemotions or various learner characteristics.

As a result of this study, instructional designers must be aware to choose the appropriate e-learningstrategy for implementing an online course, for example by reusing a pre-defined didactical componentfollowing a commonly-known learning theory. Within the scope of adult education, the three e-learningstrategies are realisable for a lecture which mainly tries to address the cognitive domain. Nevertheless,important learning objectives should be achieved by applying certain tasks which students have to masteron their own. From the didactical point of view, some kind of assessment is necessary to enforce learning.Yet, it is not important whether the assessment is done by the teacher or by the students themselves.

Drawing conclusions for this dissertation and the AdeLE prototype, the didactical strategy can be con-sidered to be relevant to online learning, whereby adaptation information addresses learner characteristicsand factors influencing the learning process and the adaptation model comprises a didactical strategy. Thisfact is also regarded by the FORMABLE model, which defines the adaptation model as an extension ofthe didactical model by inheriting from it and allowing adaptation towards the pedagogical suitability ofinstructions.

Concerning the evaluation of the AdeLE prototype and the EHELP system, this study outlines theimportance of building up an appropriate pedagogical model. In accordance to the case study depicted inthis chapter, it can be recommended that an adaptive e-learning environment and, particularly, its adapta-tion model should be designed by examining pedagogical factors within the learning process by means ofconcrete field studies.

Moreover and to put it in a nutshell, the design of an adaptation model primarily deals with observingand imitating pedagogical competencies of an e-teacher in action, whereby case studies are necessary tobuild a valid and useful adaptation model. This also resulted from the evaluation of the AdeLE system andtools like the Dynamic Background Library might provide certain teaching methods. Thus, adaptive e-learning can be equated to the implementation of pedagogical competencies with information technology.

Chapter 11

Conclusions and Outlook

“ Six by nine. Forty-two, 42. ”

[ Answer to the Ultimate Question of Life, the Universe, and Everything; Douglas Adams ]

At this point, one might be tempted to pose the question as to whether adaptive e-learning is theultimate solution to overcoming problematic aspects and shortcomings of technology-based learning and,in particular, purely virtual courses. To answer this question, this dissertation is summarised as a whole;conclusions towards the necessity and the benefits of adaptive e-learning are drawn up and an outlook isgiven.

11.1 Summary

The theoretical part of this work started with an excursion into the field of systems theory in order toimpart a comprehension and a theoretical framework for adaptation systems. On the other hand, as-pects of technology-based learning and teaching were examined by means of commonly-known learningparadigms, pedagogical issues and didactical principles. Combining these two theoretical streams, histor-ical approaches to adaptive e-learning and system types resulted from them demonstrate that this researchand development stream is not new at all. Overall, the three theory chapters introduce a formal framework,namely the FORMABLE model, to describe and evaluate adaptive behaviour in e-learning environments.

From the practical viewpoint, this dissertation dealt with standardised, adaptable courseware, as wellas the functionality of an adaptive e-learning environment. Therefore, exemplary standards and specifica-tions were highlighted, concrete requirements for supporting adaptive e-learning were established on thebasis of the FORMABLE model and e-learning specifications were inspected towards these requirements.Similarly, the following chapter examined methods and techniques for adapting the online learning pro-cess, built up an ideal adaptive e-learning environment from architectural and functional viewpoints andevaluated systems and projects in this context. Furthermore, the technical realisation of the AdeLE proto-type was described in detail, beginning with initial requirements and the planning stage up to the concreteadaptation model and a walk-through the system.

The third part of this work comprised a proof-of-concept for the AdeLE approach from three direc-tions: Firstly, the first prototype of the AdeLE research project was evaluated in terms of its usability andusefulness, whereby various weaknesses were identified and improvements were implemented. Secondly,the applicability of a Dynamic Background Library was evaluated for the AdeLE system and for adaptivee-learning in general. At this point, many benefits of a tool like this were discovered. Finally, a case studyexamined the influences of different e-learning strategies on the learning process. As a result, this study

157

158 11. Conclusions and Outlook

outlined weaknesses of AdeLE’s method of adapting the learning process and indicated the necessity aswell as a methodology to plan and build up a valid adaptation model.

11.2 Conclusions

Concerning the question introducing this chapter, it has to be stated that adaptive e-learning is not theultimate answer to solve different problems in the field of e-learning. On the one hand, there are otherapproaches like the shift to constructivistic learning, which focuses rather on collaborative elements andcommunication than on instructional design, or the provision of a highly customisable, tool-based e-learning environment like the WBT-Master, which allows teachers to implement an online course with abroad variety of didactical methods.

On the other hand, research on adaptive e-learning has already proven that many concepts and solu-tions are applicable and even beneficial in practice. Thus, many commercial products and standardisationapproaches include these aspects, whereby aspects of the macro-adaptive instructional approach are ob-viously more relevant than micro-adaptive instruction. Nevertheless, streams like Adaptive EducationalHypermedia allow the conclusion that current effort on adaptive e-learning addresses the fusion of didac-tic and content-based adaptation methods with pedagogical-driven micro-adaptation. In practice severalresearch projects, but only very few commercial products take this trend into consideration.

From the theoretical viewpoint, this dissertation can be understood as a solid base for further researchon (1) adaptation systems, for example the idea of the multi-purpose adaptive engine, (2) factors andlearner characteristics influencing learning and (3) the development of adaptation strategies for instruc-tional systems. Hence, the theoretical part of this work introduced the FORMABLE model, an attempt toformalise adaptive behaviour of e-learning environments. This formal approach essentially consists of acontent model, a pedagogical model and a didactical model. Adaptability and adaptivity is achieved byconsidering pedagogical states and extending the didactical model in terms of different adaptation meth-ods. Overall, a model, like FORMABLE, could be useful for designing and evaluating adaptive e-learningsystems, as shown throughout the whole work.

From a more practical point of view, functional requirements for e-learning standards and environ-ments were built up on the basis of the FORMABLE model. An inspection of existing standards and spec-ifications has shown that it is not possible to standardise adaptable courseware in order to support adaptivee-learning. Further, an evaluation of existing e-learning projects and solutions has demonstrated that mostlearning platforms do support adaptive e-learning partially, but many pedagogical issues have rarely beenconsidered so far. After all, the requirements for adaptable standardised courseware and adaptive featuresfor an e-learning environment served as a guideline for the development of the AdeLE system.

Yet, this prototype cannot be seen as an ideal, fully-featured adaptive e-learning environment to be ap-plied in practice, but rather as a powerful and flexible framework to realise and examine adaptive methodsand techniques. For instance, two innovative functions of AdeLE cover the application of the eye-trackingdevice and the utilisation of a Dynamic Background Library, which were shown as technically feasiblefor an e-learning environment. Moreover, the realisation of the Dynamic Background Library within theAdeLE system comprises the idea of retrieval-based instruction, which were also identified as promisingapproach by other researcher, for instance within the scope of the APOSDLE project.

Finally, three studies were carried out in order to evaluate the ideas of the AdeLE approach as well asthe concrete prototype. As a result, the following conclusions can be drawn from these studies:

• Firstly, the evaluation of an earlier version of the AdeLE system showed that the adaptation modelwas rather proprietary and the systemic performance was very poor. Beside various bug-fixes andusability improvements, the adaptation model was enhanced and the concept of learning state mod-elling was added to benefit from the possibilities of the eye-tracking device.

11.3. Outlook 159

• Secondly, a first version of a Dynamic Background Library was evaluated in order to redesign andimplement it for the AdeLE prototype. As a result, the Concept-Based Context Modeller whichwas realised as a part of the AdeLE system allows the definition and exploitation of conceptualspaces for learning content. This approach might be applicable for adapting the learning processtowards different educational scenarios, like tutorial support or the support of different learningstyles. Further, this subsystem also promotes the view of e-learning as a tool-repository.

• Thirdly, a case study outlined the influences of didactical strategies on learning, which manifeststhe necessity to adapt the teaching strategy in order to increase the effectiveness and the efficiencyof the knowledge transfer. On the other hand, this field study also indicates that cognitive sciencehas to be applied to achieve a valid and utilisable adaptation model.

Concluding this case study and this dissertation, for example by means of the FORMABLE model,adaptive e-learning can be seen as one attempt to implement didactical competencies within the e-learningsituation. Nevertheless, other promising approaches in this direction are presently available, and the use-fulness of adaptive methods has likely been proven for a few selected techniques and in particular appli-cation areas.

11.3 Outlook

In accordance with the theoretical and the practical part of this work, the research stream “adaptive e-learning” has arisen from the combination of content-based, didactical approaches and intelligent tutoringand is currently at the stage of developing and evaluating new techniques in order to enhance knowledgetransfer. Therefore, emerging disciplines and paradigms, such as information retrieval, data mining, textmining and also aspects of Web 2.0 or the Semantic Web have an impact on the field of adaptive e-learning.

In addition to these rather technological influences, cognitive science also comes up with new ap-proaches, for example the utilisation of the knowledge space theory, new methods to observe learningbehaviour or innovative pedagogical strategies. Yet, research on these recent streams is still in progressand the applicability of these ideas is proven for a rather limited scope. Thereafter, results and experiencefrom these approaches have to be formalised and realised within the adaptive engine and all the modellingsystems of an adaptive e-learning environment before being considered state-of-the-art.

In the scope of the AdeLE research project, the two major objectives – the application of the eye-tracking device as well as the utilisation of a Dynamic Background Library – were achieved, and a workingprototype (including the eye-tracking device) was successfully implemented. Despite this success story,many interesting directions for further research were identified at the end of the project:

• Firstly, it could be interesting to add further sensory systems in order to observe the learner. Sen-sors are able to measure other physiological signals like electromyogram, electrocardiogram, skinconductivity or respiration changes.

• Secondly, valid and useful instructional adaptation strategies have to be designed for the given aswell as for new pedagogical variables. As shown for the AdeLE system, the adaptation model israther prototypical and requires further field studies, dependent on the educational scenario andthe target group. Additionally, the usefulness of learner state modelling has to be evaluated inconnection with the AdeLE prototype which already utilises the eye-tracking device. Both theadaptation model and the first realisation of leaner state modelling might be improvable.

• Thirdly, efforts towards a multi-purpose adaptive engine could extend the scope of interest frome-learning to any other context dealing with information systems.

160 11. Conclusions and Outlook

• Fourthly, the modelling systems (for content and users) could also be used for other applicationscenarios. Thus, further research activities might aim to develop a multi-purpose version of thesesystems.

All in all, the AdeLE approach presents a promising outset for further research activities in this area.From the technological viewpoint the existing framework is applicable to support any of the researchstreams of adaptive e-learning because it is flexible and extensible due to the service-oriented approachand the design of all subsystems.

Bibliography

Abdi, H.: 1994, ‘A neural Network Primer’. Journal of Biological Systems (JBS) 2(3), 247–281. [Citedon page 24.]

Ackoff, R. L.: 1981, Creating the Corporate Future. New York: John Wiley & Sons. [Cited on page 13.]

AdeLE: 2006, ‘Website of the Research Project’. http://adele.fh-joanneum.at/ (2007-04-26).[Cited on pages 2, 5, 33, 65, 73 and 93.]

ADL: 2004, ‘Sharable Content Object Reference Model (SCORM) 2004 2nd Edition’. Advanced Dis-tributed Learning. http://www.adlnet.org/downloads/DownloadPage.aspx?ID=236 (2007-05-03). [Cited on pages 32, 34 and 39.]

ADL: 2007a, ‘SCORM c©’. Advanced Distributed Learning. http://www.adlnet.org/scorm/

index.aspx (2007-05-03). [Cited on pages 66 and 73.]

ADL: 2007b, ‘SCORM c©2004 3rd Edition Sample Run-Time Environment’. Advanced DistributedLearning. http://www.adlnet.org/scorm/20043ED/SRTE.aspx (2007-05-03). [Cited onpages 66 and 97.]

AICC: 1998, ‘Web-based computer-managed instruction, AGR10, CMI Subcommittee’. Aviation In-dustry CBT Committee. http://aicc.org/docs/AGRs/agr010v1.pdf (2007-05-03). [Cited onpage 72.]

AICC: 2007a, ‘Subcommittees’. Aviation Industry CBT Committee. http://www.aicc.org/pages/aicc2.htm (2007-05-03). [Cited on page 66.]

AICC: 2007b, ‘Website’. Aviation Industry CBT Committee. http://www.aicc.org/ (2007-05-03).[Cited on page 66.]

Akhras, F. N. and J. A. Self: 2000, ‘System Intelligence in Constructivist Learning’. International Journalof Artificial Intelligence in Education (IJAIED) 11, 344–376. [Cited on pages 33, 39, 50 and 53.]

Albert, D. and C. Hockemeyer: 1997, ‘Adaptive and Dynamic Hypertext Tutoring Systems Based onKnowledge Space Theory’. In: B. du Boulay and R. Mizoguchi (eds.): Artificial Intelligence in Edu-cation: Knowledge and Media in Learning Systems. Amsterdam: IOS Press, pp. 553–555. [Cited onpages 37, 42, 57 and 68.]

Albert, D. and C. Hockemeyer: 2002, ‘Applying Demand Analysis of a Set of Test Problems for Devel-oping Adaptive Courses’. In: Proceedings of the International Conference on Computers in Education.pp. 69–70. [Cited on pages 37, 57, 80, 83 and 129.]

ALEKS: 2007, ‘Website’. ALEKS Corporation. http://www.aleks.com/ (2007-05-06). [Cited onpage 131.]

161

http://adele.fh-joanneum.at/

http://www.adlnet.org/downloads/DownloadPage.aspx?ID=236

http://www.adlnet.org/scorm/index.aspx

http://www.adlnet.org/scorm/index.aspx

http://www.adlnet.org/scorm/20043ED/SRTE.aspx

http://aicc.org/docs/AGRs/agr010v1.pdf

http://www.aicc.org/pages/aicc2.htm

http://www.aicc.org/pages/aicc2.htm

http://www.aicc.org/

http://www.aleks.com/

Aleven, V., O. Popescu, and K. R. Koedinger: 2001, ‘Towards Tutorial Dialog to Support Self- Explana-tion: Adding Natural Language Understanding to a Cognitive Tutor’. In: Proceedings of the Interna-tional Conference on Artificial Intelligence in Education (AI-ED). pp. 246–255. [Cited on page 54.]

Ally, M.: 2004, ‘Foundations of Educational Theory for Online Learning’. In: T. Anderson and F. Elloumi(eds.): Theory and Practice of Online Learning. Athabasca: Athabasca University, pp. 3–31. [Cited onpage 28.]

Andriessen, J. and J. Sandberg: 1999, ‘Where is Education Heading and How About AI?’. InternationalJournal of Artificial Intelligence in Education (IJAIED) 10, 130–150. [Cited on page 50.]

ANSI: 2007, ‘Website’. American National Standards Institute. http://ansi.org/ (2007-05-03).[Cited on page 67.]

Apache: 2007, ‘Website of Torque’. Apache Software Foundation. http://db.apache.org/torque/(2007-05-04). [Cited on page 97.]

APOSDLE: 2007, ‘Website of the Research Project’. http://www.aposdle.tugraz.at/ (2007-05-03). [Cited on page 5.]

Ardito, C., M. D. Marsico, R. Lanzilotti, S. Levialdi, T. Roselli, V. Rossano, and M. Tersigni: 2004, ‘Us-ability of E-Learning Tools’. In: Proceedings of the working conference on Advanced visual interfaces.pp. 80–84. [Cited on pages 1 and 27.]

ARIADNE: 2004, ‘Website’. ARIADNE Foundation for the European Knowledge Pool. http://www.ariadne-eu.org/ (2007-05-03). [Cited on page 66.]

Ashby, W. R.: 1960, Design for a Brain: The Origin of Adaptive Behavior. London: Chapman & Hall.[Cited on pages 9 and 10.]

Astleitner, H. and J. M. Keller: 1995, ‘A Model for Motivationally Adaptive Computer-Assisted Instruc-tion’. Journal of Research on Computing in Education 27(3), 270–280. [Cited on page 32.]

Atkins, M. J.: 1993, ‘Theories of Learning and Multimedia Applications: An Overview’. Research Papersin Education 8(2), 251–271. [Cited on page 28.]

Bajraktarevic, N., W. Hall, and P. Fullick: 2003, ‘Incorporating Learning Styles in Hypermedia Environ-ment: Empirical Evaluation’. In: Proceedings of the International Conference on Adaptive Hypermediaand Adaptive Web-Based Systems (AH). pp. 41–52. [Cited on page 130.]

Banathy, B. H.: 1988, ‘Systems Inquiry in Education’. Systems Practice and Action Research 1(2), 193–212. [Cited on page 13.]

Banathy, B. H. and P. M. Jenlink: 2004, ‘Systems Inquiry and Its Application in Education’. In: D. H.Jonassen (ed.): Handbook of Research on Educational Communications and Technology. Mahwah:Lawrence Erlbaum Associates, second edition, pp. 37–58. [Cited on pages 9, 10, 11 and 12.]

Bandura, A.: 1982, ‘Self-efficacy mechanism in human agency’. American Psychologist 37(2), 122–147.[Cited on page 33.]

Bauer, M., B. Brandherm, J. Furnkranz, G. Grieser, A. Hotho, A. Jedlitschka, and A. Kroner (eds.): 2005,Lernen, Wissensentdeckung und Adaptivitat (LWA). Saarbrucken: DFKI. [Cited on page 83.]

Baumgartner, P., H. Hafele, and K. Maier-Hafele: 2002, E-Learning Praxishandbuch: Auswahl von Lern-plattformen: Marktubersicht-Funktionen-Fachbegriffe. Innsbruck: StudienVerlag. (in German). [Citedon page 90.]

162

http://ansi.org/

http://db.apache.org/torque/

http://www.aposdle.tugraz.at/

http://www.ariadne-eu.org/

http://www.ariadne-eu.org/

Benyon, D. R.: 1993, ‘Adaptive Systems: A Solution to Usability Problems’. User Modeling and User-Adapted Interaction (UMUAI) 3(1), 65–87. [Cited on page 16.]

Benyon, D. R. and D. M. Murray: 1993, ‘Adaptive Systems: From intelligent tutoring to autonomousagents’. Knowledge-Based Systems 6(4), 197–219. [Cited on page 19.]

Bergstrom, S.: 2001, CMI guidelines for interoperability, CMI Subcommittee. Aviation Industry CBTCommittee (AICC). [Cited on page 72.]

Berners-Lee, T.: 1996, ‘WWW: Past, Present, and Future’. Computer 29(10), 69–77. [Cited on page 65.]

Bhalerao, A. and A. Ward: 2001, ‘Towards Electronically Assisted Peer Assessment: A Case Study’.Association for Learning Technology journal (ALT-J) 9(1), 26–37. [Cited on page 40.]

Blaschitz, A.: 2007, ‘Persistent Database Layer fur Java’. Technische Universitat Graz. (in German).[Cited on page 97.]

Bloom, B. S.: 1956, Taxonomy of Educational Objectives: The Classification of Educational Goals. NewYork: Longman. [Cited on pages xix, 36, 37 and 42.]

Bogdanov, A.: 1984, Essays in Tektology: The General Science of Organizations. Seaside: IntersystemsPublications. (translated by G. Gorelik). [Cited on page 10.]

Bolvin, J. O. and R. Glaser: 1968, ‘Developmental Aspects of Individually Prescribed Instruction’. Au-diovisual Instruction 13, 828–831. [Cited on page 51.]

Boyd, R. and J. Apps: 1980, Redefining the Discipline of Adult Education. San Francisco: Jossey-Bass.[Cited on page 31.]

Boyle, C. and A. O. Encarnacion: 1994, ‘Metadoc: An Adaptive Hypertext Reading System’. UserModeling and User-Adapted Interaction (UMUAI) 4(1), 1–19. [Cited on page 130.]

Bransford, J. D., A. L. Brown, and R. R. Cocking: 2000, How People Learn (Expanded Edition): Brain,Mind, Experience, and School. Washington DC: National Academy Press. [Cited on pages 31, 32, 34,35, 37 and 38.]

Brennan, M.: 2003, U.S. Corporate and Government eLearning Forecast, 2002-2007. International DataCorporation. [Cited on pages 1 and 27.]

Brusilovsky, P.: 1994, ‘The Construction and Application of Student Models in Intelligent Tutoring Sys-tems’. Journal of Computer and System Sciences International 32(1), 70–89. [Cited on page 75.]

Brusilovsky, P.: 1996, ‘Methods and techniques of adaptive hypermedia’. User Modeling and User-Adapted Interaction (UMUAI) 6(2-3), 87–129. [Cited on pages 16, 23, 25, 33, 34, 79 and 80.]

Brusilovsky, P.: 2000, ‘Adaptive Hypermedia: From Intelligent Tutoring Systems to Web-Based Educa-tion’. In: Proceedings of the International Conference on Intelligent Tutoring Systems (ITS). pp. 1–7.[Cited on pages 53 and 133.]

Brusilovsky, P.: 2003a, ‘Adaptive navigation support in educational hypermedia: The role of studentknowledge level and the case for meta-adaptation’. British Journal of Educational Technology (BJET)34(4), 487–497. [Cited on page 16.]

Brusilovsky, P.: 2003b, ‘Developing adaptive educational hypermedia systems: From design models toauthoring tools’. In: T. Murray, S. Blessing, and S. Ainsworth (eds.): Authoring Tools for AdvancedTechnology Learning Environment. Norwood: Ablex, pp. 377–409. [Cited on page 118.]

163

Brusilovsky, P.: 2004a, ‘Adaptive Educational Hypermedia: From Generation to Generation’. In: Pro-ceedings of the Conference on Information and Communication Technologies in Education (ICTE). pp.19–33. [Cited on pages 1, 47 and 67.]

Brusilovsky, P.: 2004b, ‘KnowledgeTree: A Distributed Architecture for Adaptive E-Learning’. In: Pro-ceedings of the World Wide Web Conference (WWW). pp. 104–113. [Cited on pages xvii, 17, 57, 81and 134.]

Brusilovsky, P. and J. Eklund: 1998, ‘A Study of User Model Based Link Annotation in EducationalHypermedia’. Journal of Universal Computer Science (J.UCS) 4(4), 429–448. [Cited on page 130.]

Brusilovsky, P., J. Eklund, and E. Schwarz: 1998, ‘Web-Based Education for All: A Tool for DevelopmentAdaptive Courseware’. Computer Networks and ISDN Systems 30(1-7), 291–300. [Cited on pages 87and 130.]

Brusilovsky, P. and H. Nijhavan: 2002, ‘A Framework for Adaptive E-Learning Based on DistributedRe-usable Learning Activities’. In: Proceedings of the World Conference on E-Learning in Corporate,Government, Healthcare, & Higher Education (E-Learn). pp. 154–161. [Cited on page 59.]

Brusilovsky, P. and L. Pesin: 1998, ‘Adaptive Navigation Support in Educational Hypermedia: An Eval-uation of the ISIS-Tutor’. Journal of Computing and Information Technology 6(1), 27–38. [Cited onpage 130.]

Brusilovsky, P. and C. Peylo: 2003, ‘Adaptive and Intelligent Web-based Educational Systems’. Interna-tional Journal of Artificial Intelligence in Education (IJAIED) 13, 156–169. [Cited on page 2.]

Brusilovsky, P. and R. Rizzo: 2002, ‘Map-Based Horizontal Navigation in Educational Hypertext’. In:Proceedings of the Conference on Hypertext and Hypermedia (HT). pp. 1–10. [Cited on page 34.]

Brusilovsky, P., E. Schwarz, and G. Weber: 1996, ‘ELM-ART: An Intelligent Tutoring System on WorldWide Web’. In: Proceedings of the International Conference on Intelligent Tutoring Systems (ITS). pp.261–269. [Cited on page 87.]

Brusilovsky, P., S. A. Sosnovsky, and M. Yudelson: 2006, ‘Addictive Links: The Motivational Value ofAdaptive Link Annotation in Educational Hypermedia’. In: Proceedings of the International Confer-ence on Adaptive Hypermedia and Adaptive Web-Based Systems (AH). pp. 51–60. [Cited on page 131.]

C2k: 2003, ‘Website’. Classroom 2000, Northern Ireland. http://www.c2kni.org.uk/ (2007-05-03).[Cited on pages 5 and 71.]

Campbell, M., A. Goold, and P. Goward: 2004, ‘Using Online Technologies: Does Culture Matter?’. In:Proceedings of the World Conference on E-Learning in Corporate, Government, Healthcare, & HigherEducation (E-Learn). pp. 2300–2307. [Cited on pages 33 and 140.]

Campus02: 2007, ‘Website’. Campus02 University of Applied Sciences. http://www.campus02.at/(2007-04-26). [Cited on page 143.]

Carmagnola, F., F. Cena, C. Gena, and I. Torre: 2005, ‘MUSE: A Multidimensional Semantic Environmentfor Adaptive Hypermedia Systems’. In: Proceedings of the Annual Workshop of the SIG Adaptivity andUser Modeling in Interactive Systems (ABIS). pp. 14–19. [Cited on pages 70 and 73.]

Carrier, C. and D. H. Jonassen: 1988, ‘Adapting courseware to accommodate individual differences’.In: D. H. Jonassen (ed.): Instructional Design for Microcomputer Courseware. Hillsdale: LawrenceErlbaum Associates, pp. 61–96. [Cited on page 49.]

164

http://www.c2kni.org.uk/

http://www.campus02.at/

CEN: 2007, ‘Website’. European Committee for Standardization. http://www.cen.eu/cenorm/

homepage.htm (2007-05-03). [Cited on page 67.]

Chang, V. and C. Gutl: 2007, ‘E-Learning Ecosystem (ELES): A Holistic Approach for the Develop-ment of more Effective Learning Environment for Small-to-Medium Sized Enterprises (SMEs)’. In:Proceedings of the IEEE International Digital Ecosystems and Technologies Conference. [Cited onpage 13.]

Chatti, M. A., R. Klamma, C. Quix, and D. Kensche: 2005, ‘LM-DTM: An Environment for XML-Based,LIP/PAPI-Compliant Deployment, Transformation and Matching of Learner Models’. In: Proceedingsof the International Conference on Advanced Learning Technologies (ICALT). pp. 567–569. [Cited onpage 72.]

Checkland, P. B.: 1981, Systems Thinking. New York: John Wiley & Sons. [Cited on pages 12 and 14.]

Chorfi, H. and M. Jemni: 2004, ‘PERSO: Towards an Adaptive E-Learning System’. Journal of InteractiveLearning Research (JILR) 15(4), 433–447. [Cited on page 89.]

Churchman, C. W.: 1971, The Design of Inquiring Systems: Basic Concepts of Systems and Organization.New York: Basic Books. [Cited on page 14.]

Cocoon: 2007, ‘Website’. Cocoon Technologies Software AG. http://cocoontecs.org (2007-05-04). [Cited on page 90.]

Conlan, O.: 2005, ‘The Multi-Model, Metadata Driven Approach to Personalised eLearning Services’.Dissertation, University of Dublin. [Cited on pages 17, 59, 80, 84, 86 and 89.]

Cooper, P. A.: 1993, ‘Paradigm Shifts in Designed Instruction: From Behaviorism to Cognitivism toConstructivism’. Educational Technology 33(5), 12–19. [Cited on page 28.]

Corno, L. and R. E. Snow: 1986, ‘Adapting Teaching to Individual Differences Among Learners’. In:M. C. Wittrock (ed.): Handbook of Research on Teaching. New York: Macmillan, third edition, pp.605–629. [Cited on pages 2, 32, 47 and 48.]

Cronbach, L. J.: 1957, ‘Two Disciplines of Scientific Psychology’. American Psychologist 12, 671–684.[Cited on pages 32 and 48.]

DCMI: 2002, ‘DLNET: Brief Introduction to Learning Objects in DLNET’. Digital Library Network forEngineering and Technology. http://www.dlnet.vt.edu/Resources/reports/ARI_LO_Def.pdf (2007-05-03). [Cited on page 68.]

DCMI: 2007, ‘Website’. Dublin Core Metadata Initiative. http://dublincore.org/ (2007-05-03).[Cited on pages 66, 69 and 71.]

DeBra, P.: 2000, ‘Pros and Cons of Adaptive Hypermedia in Web-Based Education’. Journal on Cy-berPsychology and Behavior 3(1), 71–77. [Cited on pages 18, 53 and 87.]

DeBra, P., A. Aerts, D. Smits, and N. Stash: 2002, ‘AHA! Version 2.0: More Adaptation Flexibilityfor Authors’. In: Proceedings of the World Conference on E-Learning in Corporate, Government,Healthcare, & Higher Education (E-Learn). pp. 240–246. [Cited on page 89.]

DeBra, P., P. Brusilovsky, and G.-J. Houben: 1999a, ‘Adaptive Hypermedia: From Systems to Frame-work’. ACM Computing Surveys 31(4). [Cited on page 130.]

165

http://www.cen.eu/cenorm/homepage.htm

http://www.cen.eu/cenorm/homepage.htm

http://cocoontecs.org

http://www.dlnet.vt.edu/Resources/reports/ARI_LO_Def.pdf

http://www.dlnet.vt.edu/Resources/reports/ARI_LO_Def.pdf

http://dublincore.org/

DeBra, P., G.-J. Houben, and H. Wu: 1999b, ‘AHAM: A Dexter-based Reference Model for AdaptiveHypermedia’. In: Proceedings of the ACM Conference on Hypertext and Hypermedia (HT). pp. 147–156. [Cited on page 58.]

DeBra, P. and J.-P. Ruiter: 2001, ‘AHA! Adaptive Hypermedia for All’. In: Proceedings of the WorldConference of the WWW and Internet (WebNet). pp. 262–268. [Cited on pages 89 and 134.]

DeJong, K. A.: 1975, ‘An Analysis of the Behavior of a Class of Genetic Adaptive Systems’. Dissertation,University of Michigan. [Cited on pages 15 and 58.]

DeKoch, N. P.: 2000, ‘Software Engineering for Adaptive Hypermedia Systems: Reference Model, Mod-eling Techniques and Development Process’. Dissertation, Ludwig-Maximilians-Universitat Munich.[Cited on pages 23, 58, 59, 79 and 80.]

Deubel, P.: 2003, ‘An Investigation of Behaviorist and Cognitive Approaches to Instructional Multime-dia Design’. Journal of Educational Multimedia and Hypermedia (JEMH) 12(1), 63–90. [Cited onpage 29.]

Dietinger, T.: 2003, ‘Aspects of E-Learning Environments’. Dissertation, Graz University of Technology.[Cited on pages 1, 27, 28, 35, 38, 66, 82 and 84.]

Dietinger, T., C. Gutl, B. Knogler, D. Neussl, and K. Schmaranz: 1999, ‘Dynamic Background Libraries -New Developments in Distance Education Using HIKS (Hierarchical Interactive Knowledge System)’.Journal of Universal Computer Science (J.UCS) 5(1), 2–10. [Cited on pages 35, 53, 81, 84 and 133.]

Dietinger, T., C. Gutl, H. Maurer, M. Pivec, and K. Schmaranz: 1998, ‘Intelligent Knowledge Gatheringand Management as New Ways of an Improved Learning Process’. In: Proceedings of the WorldConference of the WWW and Internet (WebNet). pp. 244–249. [Cited on page 134.]

Dimock, K. V. and M. Boethel: 1999, Constructing Knowledge with Technology. Austin: SouthwestEducational Development Laboratory. [Cited on page 30.]

Dolog, P. and N. Henze: 2003, ‘Personalization Services for Adaptive Educational Hypermedia’. In:Proceedings of the Annual Workshop of the SIG Adaptivity and User Modeling in Interactive Systems(ABIS). [Cited on page 104.]

Dreher, H., H. Krottmaier, and H. Maurer: 2004a, ‘What we Expect from Digital Libraries’. Journal ofUniversal Computer Science (J.UCS) 10(9), 1110–1122. [Cited on page 19.]

Dreher, H., N. Scerbakov, and D. Helic: 2004b, ‘Thematic Driven Learning’. In: Proceedings of the WorldConference on E-Learning in Corporate, Government, Healthcare, & Higher Education (E-Learn). pp.2594–2600. [Cited on pages 102 and 141.]

DuBoulay, B. and R. Luckin: 2001, ‘Modelling Human Teaching Tactics and Strategies for TutoringSystems’. International Journal of Artificial Intelligence in Education (IJAIED) 12, 235–256. [Citedon page 54.]

Duffy, T. M. and D. J. Cunningham: 1996, ‘Constructivism: Implications for the Design and Deliveryof Instruction’. In: D. H. Jonassen (ed.): Handbook of Research on Educational Communications andTechnology. New York: Simon & Schuster Macmillan, pp. 170–198. [Cited on pages 29 and 30.]

Dunn, R. S., K. J. Dunn, and G. E. Price: 1989, Learning Styles Inventory. Lawrence: Price Systems.[Cited on page 33.]

166

Durand, T.: 2000, ‘Forms of Incompetence.’. In: R. Sanchez and A. Heene (eds.): Theory Developmentfor Competence-Based Management. Stanford: JAI Press, pp. 69–95. [Cited on page 36.]

Duval, E.: 2001, ‘Metadata Standards: What, Who & Why’. Journal of Universal Computer Science(J.UCS) 7(7), 591–601. [Cited on pages 65 and 66.]

Duval, E., E. Forte, K. Cardinaels, B. Verhoeven, R. V. Durm, K. Hendrikx, M. W. Forte, N. Ebel, M. Ma-cowicz, K. Warkentyne, and F. Haenni: 2001, ‘The ARIADNE Knowledge Pool System: A DistributedDigital Library for Education’. Communications of the ACM 44(5), 73–78. [Cited on pages 66 and 71.]

Ebbinghaus, H.: 1963, Memory: A Contribution to Experimental Psychology. New York: Dover. [Citedon page 34.]

Ebner, M., N. Scerbakov, and H. Maurer: 2006, ‘New Features for eLearning in Higher Education forCivil Engineering’. Journal of Universal Science and Technology of Learning (J.USTL) 0(0), 93–106.[Cited on page 88.]

EducaNext: 2007, ‘The EducaNext Portal for Learning Resources’. http://www.educanext.org/

ubp (2007-05-03). [Cited on page 71.]

Eklund, J. and K. Sinclair: 2000, ‘An empirical appraisal of adaptive interfaces for instructional systems’.Educational Technology & Society 3(4), 165–177. [Cited on page 53.]

Entwistle, N.: 1981, Styles of learning and teaching. Chichester: John Wiley & Sons. [Cited on page 31.]

EPHRAS: 2007, ‘Website of the Research Project’. http://www.ephras.org/ (2007-04-26). [Citedon page 33.]

Falmagne, J.-C., E. Cosyn, J.-P. Doignon, and N. Thiery: 2003, ‘The Assessment of Knowledge, in Theoryand in Practice’. In: Proceedings of the International Conference on Integration of Knowledge IntensiveMulti-Agent Systems (KIMAS). pp. 609–615. [Cited on pages 34, 67 and 131.]

Farzan, R. and P. Brusilovsky: 2006, ‘Social Navigation Support in a Course Recommendation System’.In: Proceedings of the International Conference on Adaptive Hypermedia and Adaptive Web-BasedSystems (AH). pp. 91–100. [Cited on page 82.]

Federico, P.-A.: 1983, ‘Changes in The Cognitive Components of Achievement as Students Proceedthrough Computer-Managed Instruction’. Journal of Computer-Based Instruction 9(4), 156–168. [Citedon page 49.]

Felder, R. M. and E. R. Henriques: 1995, ‘Learning and Teaching Styles in Foreign and Second LanguageEducation’. Foreign Language Annals 28(1), 21–31. [Cited on page 33.]

Fink, J., A. Kobsa, and A. Nill: 1998, ‘Adaptable and Adaptive Information Provision for All Users,Including Disabled and Elderly People’. New Review of Hypermedia and Multimedia 4, 163–188.[Cited on page 53.]

Fitzgerald, J. and P. G. Larsen: 1998, Modelling Systems: Practical Tools and Techniques in SoftwareDevelopment. Cambridge: Cambridge University Press. [Cited on pages 14 and 21.]

Flanagan, J. C., W. M. Shanner, H. J. Brudner, and R. W. Marker: 1975, ‘An individualized instructionalsystem: PLAN’. In: H. Talmadge (ed.): Instructional Design for Microcomputer Courseware. Berke-ley: McCutchan, pp. 136–167. [Cited on page 51.]

Fleming, M. L. and W. H. Levie: 1993, Instructional Message Design: Principles from the Behavioraland Cognitive Sciences. Englewood Cliffs: Educational Technology Publications. [Cited on page 31.]

167

http://www.educanext.org/ubp

http://www.educanext.org/ubp

http://www.ephras.org/

Francois, C.: 1999, ‘Systemics and Cybernetics in a Historical Perspective’. Systems Research and Be-havioral Science 16(3), 203–219. [Cited on page 10.]

Fritz, A., N. Galley, and C. Groetzner: 1992, ‘Zum Zusammenhang von Leistung, Aktivierung und Moti-vation bei Kindern mit unterschiedlichen Hirnfunktionsstorungen’. Zeitschrift fur Neuropsychologie 1,79–92. (in German). [Cited on page 94.]

Froschl, C.: 2005, ‘User Modeling and User Profiling in Adaptive E-learning Systems’. Master’s thesis,Graz University of Technology. [Cited on pages xvii, 99, 100 and 101.]

Galley, N.: 2001, ‘Physiologische Grundlagen, Meßmethoden und Indikatorfunktion der okulomo-torischen Aktivitat’. In: F. Rosler (ed.): Enzyklopadie der Psychologie: Grundlagen und Methodender Psychophysiologie, Vol. 4. Gottingen: Hogrefe, pp. 237–316. (in German). [Cited on page 94.]

Garcia-Barrios, V. M.: 2006, ‘Real-Time Learner Modeling: Using Gaze-Tracking in Distributed AdaptiveE-Learning Environments’. In: Proceedings of the International Convention MIPRO 2006 (CE). pp.185–190. [Cited on pages 76, 95 and 100.]

Garcia-Barrios, V. M., C. Gutl, and F. Modritscher: 2004a, ‘EHELP - Enhanced E-Learning Repositories:The Use of a Dynamic Background Library for a Better Knowledge Transfer Process’. In: Proceedingsof the International Conference on Interactive Computer Aided Learning (ICL). [Cited on pages 5, 133,135 and 136.]

Garcia-Barrios, V. M., C. Gutl, and M. Pivec: 2002, ‘Semantic Knowledge Factory: A New Way ofCognition Improvement for the Knowledge Management Process’. In: Proceedings of the InternationalConference on Society for Information Technology & Teacher Education (SITE). pp. 168–172. [Citedon pages xviii, 53, 93, 134, 135 and 136.]

Garcia-Barrios, V. M., C. Gutl, A. Preis, K. Andrews, M. Pivec, F. Modritscher, and C. Trummer: 2004b,‘AdELE: A Framework for Adaptive E-Learning through Eye Tracking’. In: Proceedings of the Inter-national Conference on Knowledge Management (I-KNOW). pp. 609–616. [Cited on page 94.]

Garcia-Barrios, V. M., F. Modritscher, and C. Gutl: 2005, ‘Personalisation versus Adaptation? A User-centred Model Approach and its Application’. In: Proceedings of the International Conference onKnowledge Management (I-KNOW). pp. 120–127. [Cited on page 20.]

Gardner, H.: 1993, Multiple Intelligences: The Theory in Practice. New York: Basic Books. [Cited onpage 32.]

Gershenson, C. and F. Heylighen: 2003, ‘When Can We Call a System Self-Organizing?’. In: Proceedingsof the European Conference on Advances in Artificial Life (ECAL). pp. 606–614. [Cited on page 18.]

Giarratano, J. C. and G. Riley: 1994, Expert Systems: Principles and Programming. Boston: PWSPublishing. [Cited on page 24.]

Glaser, R.: 1977, Adaptive Education: Individual Diversity and Learning. New York: Holt, Rinehart &Winston. [Cited on page 48.]

Google: 2007, ‘Website’. http://www.google.com/ (2007-05-04). [Cited on page 103.]

Gredler, M. E.: 2004, ‘Games and Simulations and Their Relationship to Learning’. In: D. H. Jonassen(ed.): Handbook of Research on Educational Communications and Technology. Mahwah: LawrenceErlbaum Associates, second edition, pp. 571–582. [Cited on pages 40 and 54.]

168

http://www.google.com/

Gries, V.: 2003, ‘Nutzung von Standards bei der Entwicklung von e-Learning Content’. ANOVA Multi-media Studios. http://elearning.anova.de/de/fachwissen/pdf/Contententwicklung_

Standards.pdf (2007-05-03, in German). [Cited on pages xvii, 66 and 67.]

Gutl, C.: 2002, ‘Ansatze zur modernen Wissensauffindung im Internet’. Dissertation, Technische Univer-sitat Graz. (in German). [Cited on page 103.]

Gutl, C. and V. M. Garcia-Barrios: 2005a, ‘The Application of Concepts for Learning and Teaching’. In:Proceedings of the International Conference on Interactive Computer Aided Learning (ICL). [Cited onpage 41.]

Gutl, C. and V. M. Garcia-Barrios: 2005b, ‘Towards an Advanced Modeling System applying a Service-based Approach’. In: Proceedings of the International Conference on Advanced Learning Technologies(ICALT). pp. 860–862. [Cited on pages 24, 25 and 99.]

Gutl, C., V. M. Garcia-Barrios, and F. Modritscher: 2004, ‘Adaptation in E-Learning Environmentsthrough the Service-Based Framework and its Application for AdeLE’. In: Proceedings of the WorldConference on E-Learning in Corporate, Government, Healthcare, & Higher Education (E-Learn). pp.1891–1898. [Cited on pages 5 and 140.]

Gutl, C. and F. Modritscher: 2005, ‘Towards a Generic Adaptive System applicable for Web-based Learn-ing Management Environments’. In: Proceedings of the Annual Workshop of the SIG Adaptivity andUser Modeling in Interactive Systems (ABIS). pp. 26–31. [Cited on page 5.]

Gutl, C., M. Pivec, C. Trummer, V. M. Garcia-Barrios, F. Modritscher, J. Pripfl, and M. Umgeher:2005, ‘AdeLE (Adaptive e-Learning with Eye-Tracking): Theoretical Background, System Architec-ture and Application Scenarios’. European Journal of Open, Distance and E-Learning (EURODL)II. http://www.eurodl.org/materials/contrib/2005/Christian Gutl.htm (2007-05-07). [Cited on pages 5and 94.]

Gunawardena, C. N. and M. S. McIsaac: 2004, ‘Distance Education’. In: D. H. Jonassen (ed.): Handbookof Research on Educational Communications and Technology. Mahwah: Lawrence Erlbaum Asso-ciates, second edition, pp. 355–395. [Cited on pages 27 and 39.]

Halasz, F. and M. Schwartz: 1994, ‘The Dexter Hypertext Reference Model’. Communications of theACM 37(2), 30–39. [Cited on page 58.]

Hasebrook, J. P. and H. A. Maurer: 2004, Learning Support Systems for Organizational Learning. Singa-pore: World Scientific Publishing. [Cited on pages 1, 30, 32, 33 and 36.]

Hayhoe, M. M., D. H. Ballard, J. Triesch, H. Shinoda, P. Aivar, and B. Sullivan: 2002, ‘Vision in Naturaland Virtual Environments’. In: Proceedings of the Symposium on Eye Tracking Research & Applica-tions (ETRA). pp. 7–13. [Cited on page 94.]

Helic, D.: 1995, ‘Technology-supported management of collaborative learning processes’. InternationalJournal of Learning and Change (IJLC) 1(3), 285–298. [Cited on page 38.]

Helic, D., H. Maurer, and N. Scerbacov: 2003, ‘Knowledge Transfer Processes in a Modern WBT Sys-tem’. Journal of Network and Computer Applications 27(3), 163–190. [Cited on page 88.]

Henze, N.: 2000, ‘Adaptive Hyperbooks: Adaptation for Project-Based Learning Resources’. Disserta-tion, University of Hannover. [Cited on pages 34, 80, 81, 87 and 88.]

Henze, N. and W. Nejdl: 2004, ‘A Logical Characterization of Adaptive Educational Hypermedia’. NewReview of Hypermedia and Multimedia 10, 77–113. [Cited on pages 58, 59 and 129.]

169

http://elearning.anova.de/de/fachwissen/pdf/Contententwicklung_Standards.pdf

http://elearning.anova.de/de/fachwissen/pdf/Contententwicklung_Standards.pdf

Heylighen, F.: 2003, ‘The Science of Self-organization and Adaptivity’. In: Knowledge Management,Organizational Intelligence and Learning, and Compexity, The Encyclopedia of Life Support Systems.Oxford: Eolss Publishers. [Cited on page 18.]

Hicks, D. L. and K. Tochtermann: 2001, ‘Personal Digital Libraries and Knowledge Management’. Jour-nal of Universal Computer Science (J.UCS) 7(7), 550–565. [Cited on page 25.]

Hinze, A.: 2003, ‘A-MEDIAS: Concept and Design of an Adaptive Integrating Event Notification Ser-vice’. Dissertation, Freie Universitat Berlin. [Cited on page 25.]

Hockemeyer, C. and D. Albert: 1999, ‘The Adaptive Tutoring System RATH’. In: Proceedings of theInternational Conference on Interactive Computer Aided Learning (ICL). [Cited on page 88.]

Holzinger, A., A. Pichler, and H. Maurer: 2006, ‘Multi Media e-Learning Software TRIANGLE Case-Study: Experimental Results and Lessons Learned’. Journal of Universal Science and Technology ofLearning (JUSTL) 0(0), 61–92. [Cited on page 32.]

Hooper, S. and M. J. Hannafin: 1991, ‘The effects of group composition on achievement, interaction, andlearning efficiency during computer-based cooperative instruction’. Educational Technology Researchand Development 39(3), 27–40. [Cited on pages 1 and 30.]

Hyperwave: 2007, ‘eLearning Suite’. Hyperwave AG. http://www.hyperwave.com/d/products/elearning_suite (2007-05-06). [Cited on page 136.]

iDL: 2005, ‘IDL Systems’ Adaptive Blended Learning Shows Groundbreaking Results for Adult Retrain-ing in Basic Computer Skills’. Intelligent Distance Learning Systems. http://www.idlsystems.

com/press/press_releases/press_release_march_31_2005.pdf (2007-05-06). [Cited onpage 131.]

iDL: 2007a, ‘Journal Articles’. Intelligent Distance Learning Systems. http://www.idlsystems.

com/press/journal_articles.htm (2007-05-04). [Cited on page 91.]

iDL: 2007b, ‘Press Information’. Intelligent Distance Learning Systems. http://www.idlsystems.

com/press/index.htm (2007-05-06). [Cited on page 131.]

iDL: 2007c, ‘Website’. Intelligent Distance Learning Systems. http://www.idlsystems.com/ (2007-05-04). [Cited on page 91.]

IDS: 2002, Instruction at FSU: A Guide to Teaching and Learning Practices. Instructional DevelopmentServices, Florida State University. [Cited on pages 35, 36, 37, 38 and 40.]

IEEE: 2001, ‘Public and Private Information (PAPI) for Learners (PAPI Learner)’. Institute of Electricaland Electronics Engineers. http://edutool.com/papi/drafts/08/IEEE_1484_02_01_D08_

PAPI_core_features.pdf (2007-05-03). [Cited on pages 68 and 72.]

IEEE: 2002, ‘Standard for Learning Object Metadata’. Institute of Electrical and Electronics Engi-neers. http://ltsc.ieee.org/wg12/files/LOM_1484_12_1_v1_Final_Draft.pdf (2007-05-03). [Cited on page 71.]

IEEE: 2007, ‘Learning Technology Standards Committee (LTSC)’. Institute of Electrical and ElectronicsEngineers. http://ieeeltsc.org/ (2007-05-03). [Cited on page 66.]

IICM: 2006, ‘Online Demo of the AdeLE Prototype’. Institute for Information Systems and Computer Me-dia, Graz University of Technology. http://adeledemo.iicm.tugraz.at/ (2007-05-04). [Citedon pages 99 and 109.]

170

http://www.hyperwave.com/d/products/elearning_suite

http://www.hyperwave.com/d/products/elearning_suite

http://www.idlsystems.com/press/press_releases/press_release_march_31_2005.pdf

http://www.idlsystems.com/press/press_releases/press_release_march_31_2005.pdf

http://www.idlsystems.com/press/journal_articles.htm

http://www.idlsystems.com/press/journal_articles.htm

http://www.idlsystems.com/press/index.htm

http://www.idlsystems.com/press/index.htm

http://www.idlsystems.com/

http://edutool.com/papi/drafts/08/IEEE_1484_02_01_D08_PAPI_core_features.pdf

http://edutool.com/papi/drafts/08/IEEE_1484_02_01_D08_PAPI_core_features.pdf

http://ltsc.ieee.org/wg12/files/LOM_1484_12_1_v1_Final_Draft.pdf

http://ieeeltsc.org/

http://adeledemo.iicm.tugraz.at/

IMS: 2003, ‘IMS Digital Repositories Interoperability - Core Functions Information Model’. IMSGlobal Learning Consortium. http://www.imsglobal.org/digitalrepositories/driv1p0/imsdri_infov1p0.html (2007-04-26). [Cited on page 141.]

IMS: 2007a, ‘Content Packaging Specification’. IMS Global Learning Consortium. http://www.

imsglobal.org/content/packaging/index.html (2007-05-03). [Cited on page 69.]

IMS: 2007b, ‘Contributing Members’. IMS Global Learning Consortium. http://www.imsglobal.

org/members.html (2007-05-03). [Cited on page 67.]

IMS: 2007c, ‘Learner Information Package Specification’. IMS Global Learning Consortium. http:

//www.imsglobal.org/profiles/index.html (2007-05-03). [Cited on pages 68 and 72.]

IMS: 2007d, ‘Learning Design Specification’. IMS Global Learning Consortium. http://www.

imsglobal.org/learningdesign/index.html (2007-05-03). [Cited on pages 38, 70 and 72.]

IMS: 2007e, ‘Question & Test Interoperability Specification’. IMS Global Learning Consortium. http://www.imsglobal.org/question/index.html (2007-05-03). [Cited on page 72.]

IMS: 2007f, ‘Reusable Definition of Competency or Educational Objective Specification’. IMS GlobalLearning Consortium. http://www.imsglobal.org/competencies/index.html (2007-05-03).[Cited on pages 69, 71 and 76.]

IMS: 2007g, ‘Simple Sequencing Specification’. IMS Global Learning Consortium. http://www.

imsglobal.org/simplesequencing/index.html (2007-05-03). [Cited on page 72.]

IMS: 2007h, ‘Specifications’. IMS Global Learning Consortium. http://www.imsglobal.org/

specifications.html (2007-05-03). [Cited on page 66.]

IMS: 2007i, ‘Website’. IMS Global Learning Consortium. http://www.imsglobal.org/ (2007-05-03). [Cited on page 66.]

Ivory, M. Y. and M. A. Hearst: 2001, ‘The State of the Art in Automating Usability Evaluation of UserInterfaces’. ACM Computing Surveys (CSUR) 33(4), 470–516. [Cited on page 94.]

Jain, L. C., R. J. Howlett, N. S. Ischalkaranje, and G. Tonfoni: 2002, Virtual Environments for Teaching& Learning. Singapore: World Scientific Publishing. [Cited on pages 27 and 65.]

Jameson, A.: 2001, Systems That Adapt to Their Users: An Integrative Perspective. Saarland University,Saarbrucken. [Cited on page 19.]

Jameson, A. and E. Schwarzkopf: 2002, ‘Pros and Cons of Controllability: An Empirical Study’. In:Proceedings of the International Conference on Adaptive Hypermedia and Adaptive Web-Based Systems(AH). pp. 193–202. [Cited on page 19.]

Jarvis, J. and D. Woodrow: 2001, ‘Learning preferences in relation to subjects of study of students inhigher education’. In: Proceedings of the British Educational Research Association Annual Conference.[Cited on page 33.]

John, D. and A. C. Boucouvalas: 2002, ‘The effect of the Visual-Auditory dimension of cognitive style’.In: Proceedings of the International Conference on Information Technology Interfaces (ITI). pp. 179–184. [Cited on page 130.]

Jonassen, D. H.: 2004, Handbook of Research on Educational Communications and Technology. Mahwah:Lawrence Erlbaum Associates, second edition. [Cited on page 37.]

171

http://www.imsglobal.org/digitalrepositories/driv1p0/imsdri_infov1p0.html

http://www.imsglobal.org/digitalrepositories/driv1p0/imsdri_infov1p0.html

http://www.imsglobal.org/content/packaging/index.html

http://www.imsglobal.org/content/packaging/index.html

http://www.imsglobal.org/members.html

http://www.imsglobal.org/members.html

http://www.imsglobal.org/profiles/index.html

http://www.imsglobal.org/profiles/index.html

http://www.imsglobal.org/learningdesign/index.html

http://www.imsglobal.org/learningdesign/index.html

http://www.imsglobal.org/question/index.html

http://www.imsglobal.org/question/index.html

http://www.imsglobal.org/competencies/index.html

http://www.imsglobal.org/simplesequencing/index.html

http://www.imsglobal.org/simplesequencing/index.html

http://www.imsglobal.org/specifications.html

http://www.imsglobal.org/specifications.html

http://www.imsglobal.org/

Jones, E. R.: 2002, ‘Implications of SCORMTMand Emerging E-learning Standards on Engineering Ed-ucation’. In: Proceedings of the ASEE Gulf-Southwest Annual Conference. pp. 20–22. [Cited onpage 67.]

Jones, S. and N. A. M. Maiden: 2005, ‘RESCUE: An Integrated Method for Specifying Requirementsfor Complex Sociotechnical Systems’. In: J. L. Mate and A. Silva (eds.): Requirements Engineeringfor Sociotechnical Systems. Arlington: Information Resources Press, pp. 245–265. [Cited on pages 11and 13.]

Josephson, S. and M. E. Holmes: 2002, ‘Visual Attention to Repeated Internet Images: Testing the Scan-path Theory on the World Wide Web’. In: Proceedings of the Symposium on Eye Tracking Research &Applications (ETRA). pp. 43–49. [Cited on page 94.]

Jurafsky, D. and J. H. Martin: 2000, Speech and Language Processing: An Introduction to Natural Lan-guage Processing, Computational Linguistics, and Speech Recognition. New Jersey: Prentice Hall.[Cited on page 50.]

Karampiperis, P. and D. Sampson: 2005, ‘Adaptive Learning Resources Sequencing in Educational Hy-permedia Systems’. Educational Technology Research and Development 8(4), 128–147. [Cited onpages 57 and 80.]

Kay, J.: 2000, ‘Stereotypes, Student Models and Scrutability’. In: Proceedings of the InternationalConference on Intelligent Tutoring Systems (ITS). pp. 19–30. [Cited on page 19.]

Keller, F. S.: 1968, ‘Goodbye Teacher . . . ’. Journal of Applied Behavior Analysts 1, 79–89. [Cited onpage 51.]

Keller, J. M. and K. Suzuki: 1988, ‘Use of the ARCS Motivation Model in Courseware Design’. In: D. H.Jonassen (ed.): Instructional Designs for Microcomputer Courseware. Hillsdale: Lawrence ErlbaumAssociates, pp. 401–403. [Cited on page 29.]

Kelly, D. and B. Tangney: 2003, ‘A Framework for using Multiple Intelligences in an ITS’. In: Pro-ceedings of the World Conference on Educational Multimedia, Hypermedia & Telecommunications(ED-MEDIA). pp. 2423–2430. [Cited on page 32.]

Klausmeier, H. J.: 1976, ‘Individually Guided Education: 1966-1980’. Journal of Teacher Education27(3), 199–205. [Cited on page 51.]

Knowles, M. S., E. F. Holton, and R. A. Swanson: 1998, The Adult Learner: The Definitive Classic inAdult Education and Human Resource Development. Houston: Butterworth-Heinemann, fifth edition.[Cited on page 31.]

Kolb, D. A.: 1984, Experiential Learning: Experience as the Source of Learning and Development.Englewood Cliffs: Prentice-Hall. [Cited on pages 29, 33 and 153.]

Kolbitsch, J. and H. Maurer: 2006, ‘The Transformation of the Web: How Emerging Communities Shapethe Information we Consume’. Journal of Universal Computer Science (J.UCS) 12(2), 187–213. [Citedon page 1.]

Kossiakoff, A. and W. N. Sweet: 2003, Systems Engineering Principles and Practice. Hoboken: JohnWiley & Sons. [Cited on pages 12 and 15.]

Krapp, A., S. Hidi, and K. A. Renninger: 1992, ‘Interest, Learning and Development’. In: K. A. Ren-ninger, S. Hidi, and A. Krapp (eds.): The Role of Interest in Learning and Development. Hillsdale:Lawrence Erlbaum Associates, pp. 3–26. [Cited on page 34.]

172

Kuhn, R. and S. Gudjonsdottir: 1999, ‘Virtual Campus Project: A Framework for a 3D MultimediaEducational Environment’. In: Proceedings of the Conference on Interactions in Virtual Worlds. pp.115–124. [Cited on pages 1 and 54.]

Lane, C.: 2000, ‘Learning Styles and Multiple Intelligences in Distributed Learning/IMS Projects’. TheEducation Coalition (TEC). http://www.tecweb.org/pbs/lsmidl.pdf (2007-05-03). [Cited onpage 75.]

Lennon, J. and H. Maurer: 2003, ‘Why it is Difficult to Introduce e-Learning into Schools And Some NewSolutions’. Journal of Universal Computer Science (J.UCS) 9(10), 1244–1257. [Cited on pages 1, 37,40 and 50.]

Lin, J., C. Ho, W. Sadiq, and M. E. Orlowska: 2002, ‘Using Workflow Technology to Manage Flexiblee-Learning Services’. Educational Technology & Society 5(4), 116–123. [Cited on pages 46 and 58.]

Maurer, H.: 1985, ‘Authoring systems for computer assisted instruction’. In: Proceedings of the ACMannual conference on The range of computing. pp. 551–561. [Cited on page 52.]

Maurer, H.: 1998, ‘Knowledge Management and Knowledge Transfer: Key Issues of the InformationSociety’. In: Proceedings of the European Conference on Information Society Technologies (IST). pp.132–133. [Cited on page 1.]

Mayer, R. E.: 1998, ‘Cognitive, metacognitive, and motivational aspects of problem solving’. Instruc-tional Science 26(1-2), 49–63. [Cited on page 29.]

Mayer, R. E.: 2001, Multimedia Learning. Cambridge: Cambridge University Press. [Cited on page 29.]

McLeod, G.: 2003, ‘Learning Theory and Instructional Design’. Learning Matters 2, 35–43. [Cited onpages 27 and 30.]

Modritscher, F.: 2006a, ‘e-Learning Theories in Practice: A Comparison of three Methods’. Journal ofUniversal Science and Technology of Learning (JUSTL) 0(0), 3–18. [Cited on page 5.]

Modritscher, F.: 2006b, ‘The Impact of an E-Learning Strategy on Pedagogical Aspects’. Journal ofInstructional Technology & Distance Learning (ITDL) 3(3), 85–98. [Cited on page 5.]

Modritscher, F.: 2006c, ‘Towards Formalising adaptive Behaviour within the Scope of E-Learning’. In:Proceedings of the International Conference on Adaptive Hypermedia and Adaptive Web-Based Systems(AH). pp. 362–365. [Cited on page 5.]

Modritscher, F., V. M. Garcia-Barrios, and C. Gutl: 2004a, ‘Enhancement of SCORM to support adaptiveE-Learning within the Scope of the Research Project AdeLE’. In: Proceedings of the World Conferenceon E-Learning in Corporate, Government, Healthcare, & Higher Education (E-Learn). pp. 2499–2505.[Cited on page 5.]

Modritscher, F., V. M. Garcia-Barrios, and C. Gutl: 2004b, ‘The Past, the Present and the Future ofadaptive E-Learning: An Approach within the Scope of the Research Project AdeLE’. In: Proceedingsof the International Conference on Interactive Computer Aided Learning (ICL). [Cited on page 5.]

Modritscher, F., V. M. Garcia-Barrios, C. Gutl, and D. Helic: 2006a, ‘The first AdeLE Prototype at aGlance’. In: Proceedings of the World Conference on Educational Multimedia, Hypermedia & Telecom-munications (ED-MEDIA). pp. 791–798. [Cited on pages 5, 17, 84 and 96.]

173

http://www.tecweb.org/pbs/lsmidl.pdf

Modritscher, F., V. M. Garcia-Barrios, and H. Maurer: 2005, ‘The Use of a Dynamic Background Librarywithin the Scope of adaptive e-Learning’. In: Proceedings of the World Conference on E-Learningin Corporate, Government, Healthcare, & Higher Education (E-Learn). pp. 3045–3052. [Cited onpages 5, 33, 34, 35, 81 and 128.]

Modritscher, F., C. Gutl, V. M. Garcia-Barrios, and H. Maurer: 2004c, ‘Why do Standards in the Fieldof E-Learning not fully support Learner-centred Aspects of Adaptivity?’. In: Proceedings of the WorldConference on Educational Multimedia, Hypermedia & Telecommunications (ED-MEDIA). pp. 2034–2039. [Cited on pages 5, 33 and 67.]

Modritscher, F. and A. Sindler: 2005, ‘Quizzes are not enough to reach high-level learning objectives!’. In:Proceedings of the World Conference on Educational Multimedia, Hypermedia & Telecommunications(ED-MEDIA). pp. 3275–3278. [Cited on pages 5, 31 and 35.]

Modritscher, F., S. D. Spiel, and V. M. Garcia-Barrios: 2006b, ‘Assessment in E-Learning Environments:A Comparison of three Methods’. In: Proceedings of the International Conference on Society forInformation Technology & Teacher Education (SITE). pp. 108–113. [Cited on pages 5 and 38.]

Merchant, S.: 2001, Eye Movement Research in Aviation and Commercially Available Eye Trackers Today,Eye Movement Summary. University of Iowa. [Cited on page 94.]

Merrill, M. D.: 2001, ‘Components of instruction toward a theoretical tool for instructional design’. In-structional Science 29(4-5), 291–310. [Cited on page 28.]

Microsoft: 2000, ‘Microsoft Accelerates Online Learning With Support for eLearning Specifi-cation’. http://www.microsoft.com/presspass/press/2000/Feb00/eLearningPR.mspx

(2007-05-03). [Cited on page 66.]

Miller, J. G. and J. L. Miller: 1995, ‘Applications of living systems theory’. Systems Practice and ActionResearch 8(1), 19–45. [Cited on page 12.]

Mills, B.: 2006, Theoretical Introduction to Programming. London: Springer. [Cited on page 15.]

Moen, W. E.: 1998, ‘Accessing Distributed Cultural Heritage Information’. Communications of the ACM41(4), 44–48. [Cited on page 65.]

Moodle.org: 2007, ‘Moodle - A Free, Open Source Course Management System for Online Learning’.http://moodle.org/ (2007-04-26). [Cited on pages 35, 90 and 144.]

Mozilla: 2000, ‘LiveConnect Index’. Mozilla.org. http://www.mozilla.org/js/liveconnect/

(2007-05-04). [Cited on page 98.]

Murphy, K. L. and L. Cifuentes: 2001, ‘Using Web Tools, Collaborating, and Learning Online’. DistanceEducation 22(2), 285–305. [Cited on page 30.]

Myers, I. B.: 1975, The Myers-Briggs Type Indicator (Form G). Palo Alto: Consulting PsychologistsPress. [Cited on page 29.]

MySQL: 2007, ‘Website’. MySQL AB. http://www.mysql.com/ (2007-05-04). [Cited on page 97.]

Nelson, T. and L. Narens: 1994, ‘Why Investigate Metacognition?’. In: J. Metcalfe and A. P. Shimamura(eds.): Metacognition: Knowing about Knowing. Cambridge: MIT Press, pp. 1–25. [Cited on page 33.]

NISO: 2001, ‘ANSI/NISO Z39.85: The Dublin Core Metadata Element Set’. National Information Stan-dards Organization. http://www.niso.org/standards/standard_detail.cfm?std_id=725(2007-05-03). [Cited on page 71.]

174

http://www.microsoft.com/presspass/press/2000/Feb00/eLearningPR.mspx

http://moodle.org/

http://www.mozilla.org/js/liveconnect/

http://www.mysql.com/

http://www.niso.org/standards/standard_detail.cfm?std_id=725

Noble, D.: 2001, ‘Fool´s gold: Digital diploma mills, Part V’. http://communication.ucsd.edu/dl/ddm5.html (2007-04-26). [Cited on pages 1 and 27.]

Oblinger, D. G. and B. L. Hawkins: 2005, ‘The Myth about E-Learning’. EDUCAUSE. http://www.educause.edu/ir/library/pdf/erm05411.pdf (2007-04-26). [Cited on page 27.]

Ogunnaike, B. A. and W. H. Ray: 1994, Process Dynamics, Modelling and Control. New York: OxfordUniversity Press. [Cited on page 9.]

O’Keeffe, I., A. Brady, O. Conlan, and V. Wade: 2006, ‘Just-in-Time Generation of Pedagogically Sound,Context Sensitive Personalized Learning Experiences’. International Journal on E-Learning (IJEL)5(1), 113–127. [Cited on page 89.]

OKI: 2007, ‘Website’. Open Knowledge Initiative. http://www.okiproject.org/ (2007-05-03).[Cited on page 67.]

Openwings: 2007, ‘Website’. General Dynamics C4 Systems. http://www.openwings.org/ (2007-05-04). [Cited on page 97.]

Oppermann, R.: 1994, ‘Introduction’. In: R. Oppermann (ed.): Adaptive User Support: ErgonomicDesign of Manually and Automatically Adaptable Software. Hillsdale: Lawrence Erlbaum Associates,pp. 1–13. [Cited on pages xvii and 16.]

Paepcke, A., C.-C. K. Chang, T. Winograd, and H. Garcia-Molina: 1998, ‘Interoperability for DigitalLibraries Worldwide’. Communications of the ACM 41(4), 33–43. [Cited on page 65.]

Paivio, A.: 1990, Mental Representations: A Dual Coding Approach. Oxford: Oxford University Press.[Cited on page 29.]

Park, O. and J. Lee: 2004, ‘Adaptive Instructional Systems’. In: D. H. Jonassen (ed.): Handbook ofResearch on Educational Communications and Technology. Mahwah: Lawrence Erlbaum Associates,second edition, pp. 651–684. [Cited on pages 2, 18, 24, 32, 33, 34, 35, 39, 47, 49, 52, 54, 81 and 129.]

Park, O., R. S. Perez, and R. J. Seidel: 1987, ‘Intelligent CAI: Old wine in new bottles, or a new vin-tage?’. In: G. Kearsley (ed.): Artificial Intelligence and Instruction: Applications and Methods. Boston:Addison-Wesley, pp. 11–45. [Cited on pages xvii, 1, 27 and 55.]

Paulsen, T.: 2005, ‘Emotion, memory, and stories’. In: B. Hoffman (ed.): Encyclopedia of EducationalTechnology. San Diego: San Diego State University. http://coe.sdsu.edu/eet/articles/

storyemotions/index.htm (2007-04-26). [Cited on page 32.]

Peterson, E. R., I. J. Deary, and E. J. Austin: 2003, ‘On the assessment of cognitive style: Four redherrings’. Personality and Individual Differences 34(5), 899–904. [Cited on pages 83, 123 and 127.]

Petry, B., H. Mouton, and C. M. Reigeluth: 1987, Instructional Theories in Action: Lessons IllustratingSelected Theories and Models. Hillsdale: Lawrence Erlbaum Associates. [Cited on page 36.]

Phillips, M.: 2005, ‘Cognitive Style’s influence on Media Preference: Does it matter or do they know?’. In:Proceedings of the World Conference on Educational Multimedia, Hypermedia & Telecommunications(ED-MEDIA). pp. 1023–1028. [Cited on pages 127, 131 and 153.]

Piaget, J.: 1977, The Development of Thought: Equilibration of Cognitive Structures. New York: Viking-Penguin. [Cited on page 40.]

175

http://communication.ucsd.edu/dl/ddm5.html

http://communication.ucsd.edu/dl/ddm5.html

http://www.educause.edu/ir/library/pdf/erm05411.pdf

http://www.educause.edu/ir/library/pdf/erm05411.pdf

http://www.okiproject.org/

http://www.openwings.org/

http://coe.sdsu.edu/eet/articles/storyemotions/index.htm

http://coe.sdsu.edu/eet/articles/storyemotions/index.htm

Pivec, M. and K. Baumann: 2004, ‘The Role of Adaptation and Personalisation in Classroom-BasedLearning and in e-Learning’. Journal of Universal Computer Science (J.UCS) 10(1), 73–89. [Cited onpage 25.]

Preis, A. M. and H. Mueller: 2003, ‘EyeTracking in Usability Research & Consulting: What Do theEyes Reveal about Websites & their Users’. In: Proceedings of the European Congress of Psychology:Psychology in Dialogue with Related Disciplines. p. 164. [Cited on page 94.]

Prensky, M.: 2001, Digital Game-Based Learning. New York: McGraw-Hill. [Cited on page 1.]

Pressey, S. L.: 1926, ‘A simple apparatus which gives tests and scores - and teaches’. School and Society23, 373–376. [Cited on pages 2 and 49.]

Pripfl, J., C. Trummer, and M. Pivec: 2006, ‘User Behaviour Detection by Means of Eye-Tracking’. In:Proceedings of the International Conference on Information Technology Interfaces (ITI). pp. 17–18.[Cited on page 95.]

Probst, G., S. Raub, and K. Romhardt: 2000, Managing Knowledge: Building Blocks for Success. NewYork: John Wiley & Sons. [Cited on page 1.]

Qu, C. and W. Nejdl: 2002, ‘Towards Interoperability and Reusability of Learning Resources: a SCORM-conformant Courseware for Computer Science Education’. In: Proceedings of the International Con-ference on Advanced Learning Technologies (ICALT). pp. 525–530. [Cited on pages 65 and 73.]

Rayner, K.: 1998, ‘Eye Movements in Reading and Information Processing: 20 Years of Research’.Psychological Bulletin 124(3), 372–422. [Cited on page 94.]

Reiser, R. A.: 1987, ‘Instructional Technology: A History’. In: R. M. Gagne (ed.): Instructional Technol-ogy: Foundations. Hillsdale: Lawrence Erlbaum Associates, pp. 11–48. [Cited on page 51.]

Reload: 2007, ‘Website of the Project’. Reusable ELearning Object Authoring & Delivery (RELOAD).http://www.reload.ac.uk/editor.html (2007-05-06). [Cited on page 117.]

Riding, R.: 1991, Cognitive Style Analysis. Birmingham: Learning and Training Technology. [Cited onpages 33 and 99.]

Rollet, H., M. Lux, M. Strohmaier, G. Dosinger, and K. Tochtermann: 2007, ‘The Web 2.0 way of learningwith technologies’. International Journal of Learning Technology (IJLT) 3(1), 87–107. [Cited onpage 82.]

Rollett, H.: 2003, ‘Knowledge Management Technologies in Context’. Dissertation, Graz University ofTechnology. [Cited on page 82.]

Ross, S. M. and G. R. Morrision: 1988, ‘Adapting instruction to learner performance and backgroundvariables’. In: D. H. Jonassen (ed.): Instructional Design for Microcomputer Courseware. Hillsdale:Lawrence Erlbaum Associates, pp. 227–243. [Cited on page 52.]

Rothen, W. and R. D. Tennyson: 1978, ‘Application of Bayes’ Theory in Designing Computer-BasedAdaptive Instructional Strategies’. Educational Psychologist 12, 317–323. [Cited on page 49.]

Suß, C. and B. Freitag: 2002, ‘LMML: The Learning Material Markup Language Framework’. In: Pro-ceedings of the International Conference on Interactive Computer Aided Learning (ICL). [Cited onpages 70 and 73.]

Safran, C.: 2006, ‘A Concept-Based Information Retrieval Approach for User-oriented Knowledge Trans-fer’. Master’s thesis, Graz University of Technology. [Cited on pages xvii, 102 and 103.]

176

http://www.reload.ac.uk/editor.html

Safran, C., V. M. Garcia-Barrios, and C. Gutl: 2006, ‘Concept-based Context-Modelling System for theSupport of Teaching and Learning Activities’. In: Proceedings of the International Conference onSociety for Information Technology & Teacher Education (SITE). pp. 2395–2402. [Cited on page 102.]

Sampson, D., C. Karagiannidis, and F. Cardinali: 2002a, ‘An Architecture for Web-based e-LearningPromoting Re-usable Adaptive Educational e-Content’. Educational Technology & Society 5(4), 27–37. [Cited on page 104.]

Sampson, D., C. Karagiannidis, and Kinshuk: 2002b, ‘Personalised Learning: Educational, Technologicaland Standardisation Perspective’. Interactive Educational Multimedia 4, 24–39. [Cited on pages 70and 73.]

Sanchez, J. and H. Flores: 2004, ‘AudioMath: Blind Children Learning Mathematics through Audio’. In:Proceedings of the International Conference on Disability, Virtual Reality and Associated Technologies(ICDVRAT). pp. 183–189. [Cited on page 33.]

Schmeck, R. R.: 1988, ‘Strategies and style of learning: an integration of varied perspective’. In: R. R.Schmeck (ed.): Learning strategies and learning styles. New York: Plenum Press, pp. 317–347. [Citedon page 33.]

Scouller, K.: 1998, ‘The influence of assessment method on students’ learning approaches: Multiplechoice question examination versus assignment essay’. Higher Education 35(4), 453–472. [Cited onpages 38 and 39.]

Seidel, R. J., J. G. Compton, F. F. Kopstein, R. D. Rosenblatt, and S. See: 1969, Project IMPACT; De-scription of Learning and Prescription for Instruction. Defense Technical Information Center. [Citedon page 50.]

SeLeNe: 2007, ‘Website of the Research Project’. http://www.dcs.bbk.ac.uk/selene/ (2007-05-06). [Cited on page 104.]

Shute, V. and B. Towle: 2003, ‘Adaptive E-Learning’. Educational Psychologist 38(2), 105–114. [Citedon pages xvii, 1, 47 and 56.]

Shute, V. J. and J. Psotka: 1996, ‘Intelligent Tutoring Systems: Past, Present and Future’. In: D. H.Jonassen (ed.): Handbook of Research on Educational Communications and Technology. New York:Simon & Schuster Macmillan, pp. 570–600. [Cited on page 52.]

Sindler, A.: 2005, Einfuhrung von Fernstudienelemente am Campus02. Campus02 University of AppliedSciences. (in German). [Cited on page 143.]

Singer, M.: 2005, ‘Konzeption und Implementierung eines standardkonformen Learning Content Reposi-tory’. Master’s thesis, Campus02 University of Applied Sciences. (in German). [Cited on page 71.]

Skinner, B. F.: 1974, About Behaviorism. London: Jonathan Cape. [Cited on page 28.]

Skyttner, L.: 2001, General Systems Theory: Ideas & Applications. Singapore: World Scientific Publish-ing. [Cited on page 18.]

Slavin, R. E.: 2006, Educational Psychology: Theory and Practice. Boston: Pearson Education, sixthedition. [Cited on page 32.]

Snow, E. R.: 1980, ‘Aptitude, Learner Control, and Adaptive Instruction’. Educational Psychologist 15(3),151–158. [Cited on page 48.]

177

http://www.dcs.bbk.ac.uk/selene/

Soller, A. L.: 2001, ‘Supporting Social Interaction in an Intelligent Collaborative Learning System’. In-ternational Journal of Artificial Intelligence in Education (IJAIED) 12, 40–62. [Cited on page 51.]

Sonwalkar, N.: 2002, ‘A New Methodology for Evaluation: The Pedagogical Rating of Online Courses’.Syllabus pp. 18–21. [Cited on page 131.]

Sosnovsky, S. and P. Brusilovsky: 2005, ‘Layered Evaluation of Topic-Based Adaptation to StudentKnowledge’. In: Proceedings of the Workshop on the Evaluation of Adaptive Systems, InternationalUser Modeling Conference (UM). pp. 47–56. [Cited on pages 67 and 85.]

Specht, M.: 1997, ‘Adaptive Methoden in computerbasierten Lehr/Lernssystemen’. Dissertation, Univer-sitat Trier. (in German). [Cited on page 16.]

Specht, M. and R. Oppermann: 1998, ‘ACE - Adaptive Courseware Environment’. New Review of Hyper-media and Multimedia 4, 141–161. [Cited on page 53.]

Spector, J. M. and C. Ohrazda: 2004, ‘Automating Instructional Design: Approaches and Limitations’.In: D. H. Jonassen (ed.): Handbook of Research on Educational Communications and Technology.Mahwah: Lawrence Erlbaum Associates, second edition, pp. 685–700. [Cited on pages 46 and 83.]

Stash, N., A. Cristea, and P. D. Bra: 2004, ‘Authoring of Learning Styles in Adaptive Hypermedia: Prob-lems and Solutions’. In: Proceedings of the International World Wide Web Conference (WWW). pp.114–123. [Cited on page 33.]

Stash, N., A. I. Cristea, and P. DeBra: 2006, ‘Learning Styles Adaptation Language for Adaptive Hy-permedia’. In: Proceedings of the International Conference on Adaptive Hypermedia and AdaptiveWeb-Based Systems (AH). pp. 328–332. [Cited on page 68.]

Suppes, P., J. D. Fletcher, and M. Zanotti: 1976, ‘Models of Individual Trajectories in Computer-AssistedInstruction for Deaf Students’. Journal of Educational Psychology 68(2), 117–127. [Cited on page 49.]

Sweller, J., J. J. G. VanMerrienboer, and F. G. W. C. Paas: 1998, ‘Cognitive Architecture and InstructionalDesign’. Educational Psychology Review 10(3), 251–296. [Cited on page 29.]

TALER: 2002, ‘KnowledgeTree’. Teaching And LEarning Research Lab, University of Pittsburgh. http://www.sis.pitt.edu/˜taler/KnowledgeTree.html (2007-05-06). [Cited on page 131.]

Tennyson, R. D. and D. L. Christensen: 1988, ‘MAIS: An intelligent learning System’. In: D. H. Jonassen(ed.): Instructional Design for Microcomputer Courseware. Hillsdale: Lawrence Erlbaum Associates,pp. 247–274. [Cited on page 52.]

Thorndike, E. L.: 1913, Educational Psychology, Vol. 2. New York: Columbia University Press. [Citedon page 35.]

Tobias, S.: 1987, ‘Learner Characteristics’. In: R. M. Gagne (ed.): Instructional Technology: Foundations.Hillsdale: Lawrence Erlbaum Associates, pp. 207–232. [Cited on page 32.]

Tobias, S.: 1989, ‘Another Look at Research on the Adaptation of Instruction to Student Characteristics’.Educational Psychologist 24(3), 213–227. [Cited on pages 48 and 129.]

Tobias, S.: 1994, ‘Interest, Prior Knowledge, and Learning’. Review of Educational Research 64(1),37–54. [Cited on page 32.]

Tobias, S. and T. Ingber: 1976, ‘Achievement-Treatment Interactions in Programmed Instruction’. Journalof Educational Psychology 68(1), 43–47. [Cited on page 32.]

178

http://www.sis.pitt.edu/~taler/KnowledgeTree.html

http://www.sis.pitt.edu/~taler/KnowledgeTree.html

Tochtermann, K. and G. Dittrich: 1996, ‘The Dortmund Family of Hypermedia Models: Concepts andtheir Application’. Journal of Universal Computer Science (J.UCS) 2(1), 34–56. [Cited on page 58.]

Tsypkin, Y. Z.: 1971, Adaptation and Learning in Automatic Systems. New York: Academic Press. [Citedon page 24.]

Ueno, M.: 2004, ‘Animated agent to maintain learner’s attention in e-learning’. In: Proceedings of theWorld Conference on E-Learning in Corporate, Government, Healthcare, & Higher Education (E-Learn). pp. 194–201. [Cited on page 33.]

Ulrich, W.: 1983, Critical Heuristics in Social Planning. Bern: Haupt. [Cited on page 14.]

VanDerAalst, W.: 1998, ‘The Application of Petri Nets to Workflow Management’. The Journal of Cir-cuits, Systems and Computers 8(1), 21–66. [Cited on page 65.]

Vassileva, J.: 1996, ‘A Task-Centered Approach for User Modeling in a Hypermedia Office Documen-tation System’. User Modeling and User-Adapted Interaction (UMUAI) 6(2-3), 185–224. [Cited onpage 33.]

VonBertalanffy, L.: 1956, ‘General Systems Theory’. Yearbook of the Society for General Systems Theory1, 1–10. [Cited on page 10.]

VonFoerster, H.: 1984, Observing System. Seaside: Intersystems Publications. [Cited on page 12.]

Vygotsky, L. S.: 1978, Mind in Society: The Development of Higher Psychological Processes. Cambridge:Harvard University Press. [Cited on page 50.]

Wade, V. P., H. Ashman, and B. Smyth (eds.): 2006, Adaptive Hypermedia and Adaptive Web-BasedSystems: 4th International Conference (AH 2006), Vol. 4018 of Lecture Notes in Computer Science.Berlin: Springer. [Cited on page 83.]

Wang, M. C.: 1980, ‘Adaptive Instruction: Building on Diversity’. Theory into Practice 19(2), 122–128.[Cited on page 52.]

Wasson, B. J.: 1990, ‘Determining the Focus of Instruction: Content Planning for Intelligent TutoringSystems’. Dissertation, University of Saskatchewan. [Cited on page 50.]

WBT-Master: 2007, ‘Website’. Institute for Information Systems and Computer Media, Graz Universityof Technology. http://coronet.iicm.edu/ (2007-05-04). [Cited on page 88.]

Weber, G. and M. Specht: 1997, ‘User Modeling and Adaptive Navigation Support in WWW-BasedTutoring Systems’. In: Proceedings of the International Conference on User Modeling (UM). pp. 289–300. [Cited on page 130.]

Weibelzahl, S.: 2003, ‘Evaluation of Adaptive Systems’. Dissertation, University of Trier. [Cited onpage 19.]

White, B. Y., T. A. Shimoda, and J. R. Frederiksen: 1999, ‘Enabling Students to Construct Theories ofCollaborative Inquiry and Reflective Learning: Computer Support for Metacognitive Development’.International Journal of Artificial Intelligence in Education (IJAIED) 10, 151–182. [Cited on page 51.]

Wiener, N.: 1948, Cybernetics. Cambridge: MIT Press. [Cited on pages 10 and 12.]

Wikipedia: 2007, ‘Website’. Wikimedia Foundation. http://www.wikipedia.org/ (2007-05-06).[Cited on page 103.]

179

http://coronet.iicm.edu/

http://www.wikipedia.org/

Williams, R. and H. Dreher: 2004, ‘Automatically Grading Essays with Markit c©’. Issues in InformingScience and Information Technology 1, 693–700. [Cited on pages 39 and 50.]

Winn, W. and D. Snyder: 1996, ‘Cognitive Perspectives in Psychology’. In: D. H. Jonassen (ed.): Hand-book of Research on Educational Communications and Technology. New York: Simon & SchusterMacmillan, pp. 115–122. [Cited on page 33.]

Witkin, H. A., C. A. Moore, D. R. Goodenough, and P. W. Cox: 1977, ‘Field-Dependent and Field-Independent Cognitive Styles and Their Educational Implications’. Review of Educational Research47(1), 1–64. [Cited on page 29.]

Wood, D.: 2001, ‘Scaffolding, contingent tutoring, and computer-supported learning’. International Jour-nal of Artificial Intelligence in Education (IJAIED) 12, 280–292. [Cited on page 54.]

xFIND: 2003, ‘Website of the Research Project’. Institute for Information Systems and Computer Me-dia, Graz University of Technology. http://xfind.iicm.tugraz.at/ (2007-04-26). [Cited onpage 134.]

Zhai, S.: 2003, ‘What’s in the eyes for attentive input’. Communications of the ACM 46(3), 34–39. [Citedon page 94.]

180

http://xfind.iicm.tugraz.at/

Index

Adaptability, 16Adaptable objects, 16Adaptation, 15, 16

Information, 16Model, 17Procedures, 17Rules, 16Targets, 17

Adaptation of interactive elements, 53Adaptation rules, 73Adaptation system, 15

Generic framework, 20Adaptation towards motivation, 54Adaptive collaborative e-learning, 54Adaptive component, 16Adaptive e-learning, 47

Formal models, 58Informal frameworks, 55Knowledge-based approaches, 57Logic-based models, 58Methods, 79Results and experiences, 129Techniques, 79Theoretical models, 54Workflow-based approaches, 57

Adaptive e-learning standards, 68Adaptation process, 70Didactical influences, 70Learning content, 68Pedagogical aspects, 69

Adaptive e-learning systemsArchitecture, 84Products and solutions, 90Requirements, 81, 83, 84Research projects, 88Research prototypes, 87

Adaptive engine, 16Adaptive hypermedia, 53Adaptive instructional sequening, 80Adaptivity, 16

AdeLE, 93Architectural design, 96Providing background knowledge, 95Special requirements, 93Utilising eye-tracking technology, 94

AdeLE architecture, 97Adaptive system, 98Background knowledge repository, 103Content-tracking system, 101Context modeller, 102Eye-tracking system, 101Learning management system, 97User modelling system, 99

AdeLE prototype, 103Adaptation model, 108Adaptive system, 105Evaluation study, 123Usability, 128Usefulness, 126Walkthrough, 109

Aptitude-treatment interaction, 48Artificial intelligence, 24Assessment

Problematic areas, 38Assessment methods, 39

Authoring software, 40Collaborative tasks, 40Games and simulations, 40Intelligent tutoring, 39Limited-choice question, 39Open-ended questions, 39Peer assessment, 40

Attention, 31, 154

Behaviourism, 28Bloom taxonomy, 36, 144, 151

Cognitivism, 28Collaboration, 51, 54Competency, 36

Attitude, 36

181

Knowledge, 36Skill, 36

Computer-managed instructions, 52Constructivism, 30, 50, 53Constructivistic adaptation, 53Constructivistic-collaborative approach, 50Content model, 41Content packaging, 71Context, 35Controllability, 19Courseware, 72Customisation, 19Cybernetics, 12

Data layer, 24Didactical model, 44Didactics, 35

Assessment, 38Evaluation and revision, 40Implementation, 37Planning, 36

Distributed architectural design, 24Dynamic background library, 133

Evaluation prototype, 135Evaluation study, 136Functionality, 134Necessity, 133Opportunities, 140Redesign for AdeLE, 139Utility and usability, 138

E-learning, 27Features, 82Research issues, 40

E-learning standards, 65Historical development, 66Outlook, 67Overview, 66

E-teaching, 143Case study, 143Comparison of three strategies, 145Didactical findings, 150Impact on learning, 153Influences on learners, 152

Education, 31Emotions, 32, 154Eye-tracking, 94

Factors influencing learning, 31, 34Attention, 31, 154

Context, 35Emotions, 32, 154Feedback and tutoring, 35, 155Interest, 34Motivation, 31, 155Prior knowledge, 32, 155Remembering and forgetting, 34, 155Time on-task, 34

Feedback and tutoring, 35, 155FORMABLE, 60, 68–70, 80, 81, 83

Adaptation model, 61Content model, 41Didactical model, 44Pedagogical model, 43

Formal systemic model, 15

Games and simulations, 53

Hard-systems science, 12Human systems, 12

Instruction-based adaptation techniques, 79Instructional design, 37Intelligent tutoring systems, 52Interactive communication, 50Interest, 34

Knowledge space theory, 37, 57KnowledgeTree architecture, 57

Learner characteristics, 32, 152Background knowledge, 33, 153Cognitive and learning styles, 33, 153Constitution, 33, 153Intellectual capabilities, 32Learning preferences, 33, 153Prior knowledge and experience, 34, 152Self-efficacy and meta-cognition, 33

Learner control, 48Learner profiles, 72Learning, 31Learning objectives, 36, 151Learning theories, 27

Macro-adaptive instruction, 48Macro-adaptive instructional systems, 51Meta-adaptation, 16Meta-adaptation system, 17Meta-cognition, 51Micro-adaptive instruction, 49

Diagnosic process, 49

182

Prescriptive process, 49Theoretical model, 49

Motivation, 31, 50, 54, 155Multi-purpose adaptive system, 21

Openwings, 104

Pedagogical model, 43Personalisation, 19

Dimensions, 20Prior knowledge, 32, 155Problem situations, 13

Remembering and forgetting, 34, 155Response sensitivy, 49Retrieval-based instruction, 73, 74, 81, 83, 103, 108,

124

Scrutability, 19Semantic tagging editor, 75Service-oriented approach, 24Standardised courseware for AdeLE, 73Systemic characteristics, 18

Complexity, 18Controllability, 18Feedback, 18Feedforward, 19Intelligence, 18Learnability, 18Observability, 18Openness, 18Purposiveness, 18Self-organisation, 18

Systems design, 14Systems methodology, 11Systems philosophy, 11Systems theory, 9

Basics, 10History, 10

Systems thinking, 12

Teaching, 35Time on-task, 34

User adaptation, 19

183

implementation and evaluation of pedagogical strategies in adaptive e-learning ... ·...

Documents