Implicit Feedback Based Recommendation andCollaboration

Martin Labaj∗

Institute of Informatics and Software EngineeringFaculty of Informatics and Information Technologies

Slovak University of Technology in BratislavaIlkovicova 3, 842 16 Bratislava, Slovakia

labaj@fiit.stuba.sk

AbstractRecommendation, collaboration and other tasks play im-portant role on the adaptive web. For such tasks, userfeedback is needed. Explicit feedback interrupts user andobtaining quality explicit feedback is problematic. Onthe other hand, users provide implicit feedback uninter-rupted without knowing that they are rating. Traditionalimplicit feedback on the web – tracking of mouse and key-board interaction, displayed parts of document, etc. – isproblematic, when user passively reads the document anddoes not provide any inputs. We do not want to force himto provide inputs; therefore we have to track him physi-cally. In our work we proposed a method for identificationof important fragments based on implicit interest indica-tors with included commodity gaze tracking in commonsettings of the user’s home. We use collected informationin recommendation of fragments, adaptive explicit feed-back collection and we proposed additional scenarios.

Categories and Subject DescriptorsH.4 [Information Systems Applications]: Miscella-neous; K.3.1 [Computer Uses in Education]: Collab-orative learning, Computer-assisted instruction (CAI)

Keywordsimplicit feedback, gaze tracking, adaptive web systems,TEL systems

1. Feedback on the Adaptive WebTracking of user actions is needed for many tasks on theadaptive web, including recommendation. This tracking

∗Master degree study programme in field Software Engi-neering. Supervisor: Prof. Maria Bielikova, Institute ofInformatics and Software Engineering, Faculty of Infor-matics and Information Technologies, STU in Bratislava.c© Copyright 2011. All rights reserved. Permission to make digital

or hard copies of part or all of this work for personal or classroom useis granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies show this notice onthe first page or initial screen of a display along with the full citation.Copyrights for components of this work owned by others than ACMmust be honored. Abstracting with credit is permitted. To copy other-wise, to republish, to post on servers, to redistribute to lists, or to useany component of this work in other works requires prior specific per-mission and/or a fee. Permissions may be requested from STU Press,Vazovova 5, 811 07 Bratislava, Slovakia.Labaj, M. Implicit Feedback Based Recommendation and Collabora-tion. Information Sciences and Technologies Bulletin of the ACM Slo-vakia, Special Section on the ACM Student Project of the Year 2011Competition, Vol. 3, No. 4 (2011) 41-42

is common on the server-side or at the middle-man (e.g.Adaptive Proxy [1]), but only actions such as documentload are collected, considering the document as a whole.With addition of client-side code that is reporting to theserver or proxy, user interaction within a web page is com-monly tracked. Read wear [4] can be employed this way –a virtual wearing of document while it is being displayed,essentially tracking time on screen for its fragments. Thisapproach fails if the content is short. Content is com-monly placed “above the fold” and application tools arevisible entire time. Mouse interaction provides anotherinformation and even correlates with gaze to some de-gree [3]. However, when user does not move the mouse orpress keys, we cannot differentiate if he is reading or hasleft. For example, text away from cursor can be made un-readable [7] and user has to move mouse to read. Anotherapproach is to track the user physically: detect his pres-ence with a camera or even estimate his gaze. Apart fromcostly professional devices used for evaluations in usabilitylaboratories, commodity gaze tracking using cheap web-cams has been used, e.g. in assistance to the disabled [6].

2. Gathering the Feedback and RecommendingWe proposed a method for mapping implicit feedback con-sisting of several interest indicators to importance of frag-ments of a document and of a web application presentingthe document. We divided the indicators into groups andexperimentally assigned weights between the groups andbetween indicators in the groups. Importance of a frag-ment is calculated as a product of values of all indicatorsbelonging to that fragment and their weights. Metricsof indicators vary for each type of an indicator. For ex-ample, gaze position estimated from user’s image carriessome inaccuracy and therefore it is accumulated not onlyfor the estimated fragment, but also for nearby fragmentswith value decreasing with distance. On the other hand,an annotation is precisely targeted by the user.

We then use the importance of fragments in a recommen-dation. When user first visits a document, we highlightfragments that other users have worked with. User canquickly scan through the document when he is learningor browsing in an exploratory way. When revisiting, wecan use user’s own feedback exclusively and he can revivewhat he read previously. In tracking of application frag-ments, we can differentiate whether user has not workedwith a part of application (widget) because he does notwant to (he finds selected items, scores, etc. uninterest-ing) or he has not noticed it at all. We point such userto application fragments he have not worked with yet and

42 Labaj, M.: Implicit Feedback Based Recommendation and Collaboration

Administrácia Debug SI C Martin Labaj (administrátor) | Odhlásiť

Odporúčame pozrieť:

Operace bitového posunu doprava

Spracovanie programu 2

5.3 Operátor čárky

Knižnica pre I/O

Bitový součin

Koncepty

blok , definícia premennej , lokálnapremenná , globálna premenná , premenná ,

deklarácia premennej

Texty Otázky Cvičenia

1 Úvod

2 Základní pojmy

3 První začátky s C

3.1 Jednoduché datové typy a přiřazení

Definice proměnných

Přiřazení

3.2 Hlavní program

3.3 Konstanty

Celočíselné konstanty

Reálné konstanty

Znakové konstanty

Tvoje skóre

13.9V priebežnom hodnotení sú158 študenti pred Tebou!

Nahlásené chyby

Martin Polakovic2010-12-04 17:20boxy sa prekryvaju (a uz tretikratnahlasujem chybu ...

Marek Jurena2011-02-06 16:38chyba medzera

Externé zdroje

Premenne

Programování v jazyce C - proměnné, funkce pri...

Programátor - Kurz C/C++ (2.)

Ukáž ďalšie (2)Pridaj externý zdroj

Tagy

moje populárne

K tomuto obsahu nie sú kdispozícii žiadne tvoje tagy.

Pridaj tag

?Definice proměnných

Pod pojmem definice se míní příkaz, který přidělí proměnné určitého typu jméno apaměť.

Naopak deklarace je příkaz, který pouze udává typ proměnné a její jméno. Deklaracenepřiděluje žádnou paměť! Smysl deklarací a jejich použití bude vysvětlen v kap.9.2.5.

Pozor:

V některé literatuře jsou významy slov deklarace a definice právě opačné. Přinejasnostech je tedy třeba zjistit na příkladech, co má autor na mysli.

V C jsou definice v obráceném pořadí než u Pascalu:

Pascal C

VAR i : INTEGER; c, ch : CHAR; f, g : REAL;

int i;char c, ch;float f, g;

Poznámka:

Definice proměnných se mohou vyskytnout buď vně (globální proměnná) nebo

uvnitř funkce (lokální proměnná)[1], např.:

int i; /* globalni promenna */int main(){ int j; /* lokalni promenna */}

Štábní kultura:

Filter:

Figure 1: Evaluation within an instance of ALEF (left), UTrack extension (middle), GazerTracker (right).

even adaptively collect explicit feedback and ask him whathe thinks of the application’s tool, when he noticed it sev-eral times but did not use it. If many users fail to noticea feature that may sign a problem in web design.

In order to collect the feedback from users, we imple-mented an UTrack extension (Figure 1) of Firefox webbrowser. The extension tracks user actions commonly vis-ible to web sites such as mouse interaction. With moreaccess to browser, we also track interaction with browseroutside the actual page. We implemented desktop appli-cation GazerTracker using modified library from researchproject OpenGazer. GazerTracker uses sockets both forcontrol and gaze communication with other applications.GazeTracker is automatically launched and controlled bythe UTrack extension. Collected data are mapped to vis-ited web pages. If a web application can use the informa-tion, it defines its user and application interfaces – whatconstitutes fragments and where to send processed data.

3. Evaluation and ConclusionsWe integrated UTrack with Adaptive LEarning Frame-work (ALEF) [2] developed and used on FIIT. Group ofstudents used ALEF in the Principles of software engi-neering course while preparing for small exams from pre-selected set of learning objects. Eleven students usedstandalone webcams, twelve used laptop webcams andothers used the extension without gaze tracking. Studentsalso manually highlighted parts of learning texts (docu-ment fragments) they considered important and they wereadaptively asked about they work with widgets (applica-tion fragments).

In our work we proposed a method for identification of im-portant fragments using implicit feedback including gazetracking and used it in a recommendation. Our experi-ments have shown that it is possible to bring gaze trackingto common settings of unsupervised users working withweb applications. Accuracy is comparable to sizes of typ-ical page elements (menu width, widgets). We found thatusing our method we can more accurately track interestin fragments of documents than when using only mouseinteraction and that we reflect important fragments moreaccurately than users’ intentional highlights. Users weremore willing to provide explicit feedback when we askedadaptively when we determined they were working with

concerned application fragment (6.96 % refused answers)than when we showed them based only on mouse interac-tion (12.44 % refused) and when we showed questions ran-domly (33.33 % refused). We also proposed more scenar-ios: augmentation of the communication between usersof the same or similar websites [5] or estimation of users’interests while browsing the digital space. More accu-rately tracking the user actions including gaze trackingoutside of usability laboratories in the common workingenvironment of the user brings some concerns (fear forprivacy) and problems (lower accuracy and comfort thanwith laboratory equipment), however it brings advantagesnot only to fragment identification and recommendation,but also possibly to many other areas in web environment,e.g. navigation support, context recognition.

Acknowledgements. This work was partially supportedby the Cultural and Educational Grant Agency of theSlovak Republic, grant No. KEGA 028-025STU-4/2010.

References[1] M. Barla and M. Bieliková. Ordinary web pages as a source for

metadata acquisition for open corpus user modeling. WWWInternet2010, pages 227–233, 2010.

[2] M. Bieliková, M. Šimko, M. Barla, D. Chudá, P. Michlík,M. Labaj, V. Mihál, and M. Uncík. ALEF: Web 2.0 principles inLearning and Collaboration, pages 54–49. 2010.

[3] M. C. Chen, J. R. Anderson, and M. H. Sohn. What can a mousecursor tell us more?: correlation of eye/mouse movements on webbrowsing. In CHI ’01 extended abstracts on Human factors incomputing systems, CHI EA ’01, pages 281–282, New York, NY,USA, 2001. ACM.

[4] W. C. Hill, J. D. Hollan, D. Wroblewski, and T. McCandless. Editwear and read wear. In Proceedings of the SIGCHI conference onHuman factors in computing systems, CHI ’92, pages 3–9, NewYork, NY, USA, 1992. ACM.

[5] M. Labaj. Web-based learning support based on implicit feedback.Information Sciences and Technologies Bulletin of the ACMSlovakia, 3(2):76–78, 2011.

[6] J. San Agustin, H. Skovsgaard, E. Mollenbach, M. Barret, M. Tall,D. W. Hansen, and J. P. Hansen. Evaluation of a low-costopen-source gaze tracker. In Proceedings of the 2010 Symposiumon Eye-Tracking Research & Applications, ETRA ’10, pages77–80, New York, NY, USA, 2010. ACM.

[7] C. Ullrich and E. Melis. The poor man’s eyetracker tool ofactivemath. In Proceedings of the World Conference on E-Learningin Corporate, Government, Healthcare, and Higher Education(eLearn-2002), pages 2313–2316.