the kiwi vision

ReasoningWeb Summer School 2008

Venice, September 2008

Dr. Sebastian Schaffert

Salzburg Research Forschungsgesellschaft

[email protected]

http://www.kiwi-project.eu http://planet.kiwi-project.eu

KiWi Vision

ReasoningWeb Summer School, Venice, © 2008,Sebastian Schaffert, Salzburg Research

Knowledge Management

| many different kinds of rich content

(text, images, audio, video, software, processes, …)

| user and domain specific workflows and processes

| sharing of content and collaboration of users

KiW

i Vis

ion


Knowledge Management (traditional)

| „knowledge acquisition systems“

| form-based, predefined processes, part of quality

management, „make people replaceable“

| people are aligned with technology and organisation

KiW

i Vis

ion


Knowledge Management (KiWi Way)

| instead: technology and organisation should be alignable

with people!

| KiWi: Semantic CMS the Wiki-Way

KiW

i Vis

ion


“share, give it away, make it easy, because the more

people know your idea the more powerful it becomes”

– Garr Reynolds, Presentation Zen


Online Communities

… are successful in

sharing even now

… there is no reason

why this shouldn‘t work in the enterprise


Knowledge Management (Wikis)

| Wikis are...

| simple to use (low technological barrier)

| flexible: from a short notice over documentation to collaborative authoring of documents

| do not impose a predefined workflow (no dictate of the system)

| adjust to the necessities of users

KiW

i Vis

ion


Wikis: like a piece of paper!

… you can write on it

… you can draw on it

… you can

connect things


Wikis: like a piece of paper

| workflows only by “social convention”

| there are rules, but it is possible to deviate from them if necessary (new situations, better solutions, …)


Knowledge Management (Wikis)

| but: Wikis are rather like an empty piece of paper

| well suited for creative and/or well-known tasks

| no support whatsoever for users

| nobody would fill his tax return on an empty piece of paper!

| forms and workflows have (originally) been developed as support!

| with growing amount of content it becomes also

increasingly difficult to find the necessary information

KiW

i Vis

ion


Knowledge Management + Wiki-Philosophy

+ Semantic Web = KiWi

| machine readable linking of content

| adaption of presentation and input

| to personal preferences

| to user and content context

| to different kinds of content

| examples:

| kinds of content: meeting minutes, resource plans, persons, tasks, reports, ideas, ...

| presentation/input: meeting minute editor, gantt diagram, user

profile, report template, ...

KiW

i Vis

ion


KiWi Use Cases: Requirements

KiW

i U

se C

ase

s


KiW

i U

se C

ase

s

Software Knowledge Management

Sun Microsystems (Prague)

| knowledge in and about software systems involves many different kinds of (rich) content:

| source code

| documentation

| minutes of group meetings

| bug reports,

| ...

| development of Sun’s Netbeans IDE is developed by a large community with a team of 150 core developers employed by Sun Prague

| knowledge gathered and developed during the software development process is distributed over many different kinds of systems (wikis, bug trackers, blogs, discussion forums, mailing lists)

KiW

i U

se C

ase

s



Sun Microsystems (Prague)

| goal: to develop a platform for managing and sharing knowledge in virtual software development communities

KiW

i U

se C

ase

s



Pilot Roles

| designer: responsible for requirements and use cases, usability (user researcher), as well as system behaviour (interaction designer) and user interface (visual designer)

| developer: responsible for actual implementation of planned features defined by planners and further specified by designers

KiW

i U

se C

ase

s



Designer – KM Tasks

| write UI specification and enter related bugs to the bug repository

| understand high-level relationships between various documents

| relate document or its part to an existing concept

| define non-existing entity while writing a text document

KiW

i U

se C

ase

s


Project Knowledge Management

Logica Denmark

| Logica DK: provider of IT solutions and services to public

and private sector

| currently undergoes CMMI ML2 certification for their project

management processes as a requirement by customers

KiW

i U

se C

ase

s



CMMI ML 2

| CMMI ML2:

“Capability Maturity Model Integration Maturity Level 2”

means:

| standardising work processes among projects

| documenting and managing work processes

| following certain workflows

| sharing process knowledge between projects

KiW

i U

se C

ase

s



Logica Denmark

| goal: to develop a platform documenting and supporting processes in project management conforming to CMMI ML 2 and exchange relevant knowledge

KiW

i U

se C

ase

s



Roles

| process engineer: works on process for the company in

general and teams in particular; all process changes have to

be acknowledged by process engineer

| project manager: mediator between customer and the

team; discusses project, implementation, and goals with

both, customer and team members

| team member: software developers, architects, testers,

etc.; experienced in their respective tools and knows his

peers

KiW

i U

se C

ase

s



Required KiWi Features

| collaborative environment where knowledge and experience

can be easily written both at the project and process level

| support for semantic tagging and annotation with semantic

meta-data

| support for semantic search over process descriptions,

project documentation, and reported experience

| support for identification of well-defined inconsistencies

between selected work products

KiW

i U

se C

ase

s


KiWi Photo Stories

KiW

i U

se C

ase

s


KiWi Concepts

KiW

i Core

Conce

pts


Core Usage Concepts

| content items: single units of information, associated with

a URI (for machine-readable data) and some content

(human-readable)

| tags: textual annotations of content items; can be lifted to

semantic concepts, thus embedding content items in a

knowledge base

| triples: relate two content items; stored in RDF; predicates

may be further specified in ontologies

KiW

i Core

Conce

pts


Core Usage Concepts

| users: play an important role for personalisation, social

networking, and reputation; have a unique URI and thus a

content item

| roles: in every social interaction, users take specific roles

which they can explicitly switch (e.g. “at home”, “at work”)

| revisions: represent logical changes to the system and are

explicitly committed by the user (by clicking “save”); a

revision may contain of several atomic updates

KiW

i Core

Conce

pts


Advanced Usage Concepts

| rules: define rule-based deductive knowledge; rules can

work on content items, tags, and triples and can add

additional triples as well as influence the layout and

presentation of the content

| structured tags: allow to define more complex tagging

structures, e.g. hotel(3stars), author(sebastian), etc.

KiW

i Adva

nce

d C

once

pts


KiWi User Interface

KiW

i U

ser

Inte

rface


Presenting Content Items

| principle: content items are presented in a way that is as

suitable as possible for the user

| generic presentation:

| wiki page (known from other wikis)

| incoming/outgoing references (known from IkeWiki)

| tags

| triple context (graph visualisation)

| specific presentation:

| custom layouts and widgets for certain kinds of content

items that display the content in a manner appropriate for the

domain and personal preferences


Widgets & Layouts

| Task 1: allow the adaptation of the presentation and

editing of content to current context and personal

preferences

Layouts

| Task 2: allow (advanced) users to define custom

components („widgets“) for presenting and editing content

(and meta-data!)

Widgets

| Requirement:

| allow rules and reasoning to influence the layout using user-

defined rules


Custom Layouts & Widgets context specific content layout


Custom Layouts & Widgets context specific visualisations


Editing Content Items

| principle: allow everyone to enter data as she wants

| simple editor for simple tasks

| advanced access to the system if so desired

| guide users from very simple but restricted to full access


Annotations Editor

| generic way of annotating content items

| the annotations editor is a native RDF editor that builds

directly on top of the RDF model!

| allows to associate wiki pages (i.e. content items) and links

between pages with types

| generic and flexible, but (as experience with IkeWiki shows)

too complicated for most users!


Annotations Editor


Semantic Forms

| application and context specific forms for entering the

relevant data, supporting the user e.g. by drop down

selections, range and type restrictions, …

| selected based on personalisation/context adaptation (e.g.:

when displaying a user profile, provide a user profile form)

| may allow to change both, content and meta-data

(depending on processing instructions)

| defined by (advanced) users using widgets

(actually: a special kind of widget that allows editing data)


Semantic Forms (Slide: Denny Vrandecic / ACTIVE project)


Semantic Forms

| arguably the most simple and user-centred way of editing

content and meta-data

| requires a priori definition of forms

| mostly specific to the application domain


Browsing Content Items

| two ways of finding relevant content

| search (active)

| navigation (passive)

| search in KiWi:

| always start with a keyword or keywords, used for full-text as

well as semantic search (over RDF, Tags)

| refine search results via facetted browsing, adding additional

concepts, relations, tags

| navigation in KiWi:

| incoming/outgoing references for a content item

| navigation ontology that is rendered appropriately


KiWi Technology

KiW

i Te

chnolo

gy


Technology Stack and Frameworks (Server)

| Java EE 5: provides core technology for persistence and

data management

| JBoss AS: Java EE web and application server, supporting

clustering, persistence, …

| Sesame 2: RDF triple store supporting easy extension with

custom reasoners

| JBoss Seam: web application framework building on and

extending Java EE; provides core web application

functionalities (user management, MVC separation &

inversion of control, data access, …), Lucene integration,

allows easy extension with e.g. Web Services

| TestNG: advanced unit testing environment for web

applications


Technology Stack and Frameworks (Client)

| JSF/RichFaces: provides core user interface concepts for

web applications and supports interaction with server using

AJAX

| XHTML: for rendering presentation in the browser and for

storing conent item content

| TinyMCE: as extensible browser-based WYSIWYG editor for

XHTML


Model: Persistence & Data Model

| object oriented model for accessing frequently used core

concepts of the system (“content item”, “user”, “role”, …)

| objects mapped partly to a relational database (using JPA/

Hibernate) and partly to the triple store (Sesame);

developers don’t need to care


Controller: Seam, EJB, and the KiWi API

Controller Services (1)

| unified access to objects of the data model, regardless

how they are persisted

| direct access to the XML and RDF data of content items

and meta-data

| querying facilities for textual and meta-data querying,

later extended by the KiWi reasoning and querying

language

| web service endpoint for accessing the system from

external applications


Controller: Seam, EJB, and the KiWi API

Controller Services (2)

| widget execution environment: an environment that

provides relevant functionalities to widgets, e.g. the current

context, the data model, etc.

| personalisation: management of the user model and

automatic selection of appropriate layouts and widgets

| information extraction: allows automatic analysis of

textual content for tag and annotation recommendation

| reasoning: allows the management and execution of

predefined or user-defined rules that work on both, the

content and the meta-data


View: the KiWi user interface

| browser-based, i.e. XHTML + JavaScript

| implemented as a specific configuration of layouts and

widgets for different applications

| widgets may be implemented in different languages, e.g.

JSF, Groovy, XHTML, JavaScript, …



KiWi Research

KiW

i Rese

arc

h

KiWi Enabling Technologies

| Reasoning: a rule-based language capable of working with

both data (content items) and meta-data (triples, tags)

| Reason Maintenance: tracking justifications for

derivations for presentation to the user and for efficiency

| Information Extraction: supporting the user by

suggesting appropriate annotations based on human-

readable content

| Personalisation: adapting the presentation to the user

based on context and user model


Reasoning

Motivation

Reaso

nin

g


what should reasoning in such a setting look like?

Reasoning

Motivation

| … consistency checking?

| … instance checking?

(this is what the current Semantic Web Reasoning offers)

Reaso

nin

g


Reasoning

Motivation

or rather …

| … representing rule knowledge (processes, workflows)

| when something is tagged “bug” it should also be tagged with

“todo”

| … personalising user experience?

| list of all pages tagged “todo” that are relevant to me

| … defining reactive behaviour?

| when someone clicks on this button, add that tag

| … defining policies?

| Piero can give you the best examples

(this is what rule-based reasoning could offer)

Reaso

nin

g


Rule-based Reasoning

Motivation

| semantic wikis require different support for reasoning and

querying

| rule-based instead of DL-based

| deviations instead of consistency checking

| user-defined rules, e.g. for personalisation

| current systems (like Jena and Sesame) do not offer these

kinds of reasoning and are slow due to complex DL

reasoning tasks

Reaso

nin

g


Reason Maintenance

Motivation

| if reasoning is supposed to be used for personalisation and

recommendations …

| users must be able to get explanations and justifications for

what is recommended and personalised

| users must be able to manually override / disable certain rules

| reason maintenance is the science about how to address

these two issues, e.g. by tracking the justifications for each

derived fact

| it also helps to increase efficiency, particularly for updates,

which is essential for wiki-like systems

| it is also useful for versioning

Reaso

nin

g


Reasoning & Reason Maintenance

Goals

| to develop a rule-based language that

| can be used by users of the wiki to query and to specify

derivation (and possible action) rules

| capable to query content and knowledge base in a unified

fashion

| with a simple and intuitive way to specify such queries and

rules

| to develop a reason maintenance component for this

language that

| gives users a way to understand why certain

derivations exist,

| allows versioning of updates to the knowledge

base

| allows easy updates

Reaso

nin

g


Information extraction

Motivation

| current semantic wikis provide no support for annotating content

(besides providing an easy-to-use interface to enter annotations)

| content of wiki pages usually already contains much of the

knowledge, albeit in natural language text

| annotation is for many users a very daunting task, because it

requires understanding of the underlying knowledge models and

concepts behind them

| existing natural language technologies can partly extract meta-data

out of the natural language text and use this information to

interactively guide users in the annotation

process (e.g. by providing custom wizards, or

even by simple reordering of the offered

concepts)

Info

rmation E

xtr

act

ion


Information extraction

Goals

| to develop an information extraction component that semi-

automatically extracts meta-data out of wiki pages to

interactively guide the user through the annotation task

Info

rmation E

xtr

act

ion


Personalisation

Motivation

| different users have different roles in knowledge management

processes (e.g. developer, tester, documentation writer, customer)

| different users might also have different tasks and/or preferences

| in both cases, personalised presentation of content and user

interface provides support for the user

| a semantic wiki can support this by storing user/group models in

the knowledge base, by offering a reasoning component, and by

offering advanced visualisations and editors that implement

personalised access to the wiki

Pers

onalis

ation


Personalisation

Goals

| develop appropriate user and group models for representing

properties and preferences of groups wrt. the KIWI system

| automatically track the usage of single users and user groups in

order to refine user and group models

| develop rules that dynamically adapt the presentation of the

content and user interface to the user

Pers

onalis

ation



What I believe made KIWI a success (Scientific & Technological Content)

| KIWI has a concrete, graspable, clearly defined outcome

| at the end of the project, we will have a single, demonstrable system that actually implements what we promise

| theory touching the ground!

| KIWI addresses real-world problems identified by its industrial partners

| no blue sky research, application oriented

| KIWI improves already existing technology where it has deficiencies

| IkeWiki

| KIWI builds upon previous EU and national projects and takes their outcomes to the next level

| REWERSE, QVIZ, Dynamont


What I believe made KIWI a success (Project Setup)

| The KIWI consortium is well balanced

| complementary expertise, exactly as needed for the project

| 3 universities, 1 research centre, 3 industrial partners (1 SME)

| 4 countries (Austria, Germany, Czech Republic, Denmark)

| Significant effort goes into Dissemination & Demonstration

| 64 out of 388 PM (= 18,5%)

| the importance of marketing research results is often

underestimated

| The project workplan and budget was planned as if it was

already a description of work

| thorough and realistic calculation


What I believe made KIWI a success (Social Dynamics)

| We had a proposal kick-off meeting with all partners 2 months before the submission deadline

| people learn to know and trust each other

| a rough project workplan was set up (Gantt diagram)

| social bonds are important!

| I tried to collect issues and send out email to the partners only once a week (“KIWI weekly update”)

| no email overload at the partners (be nice to them!)

| clear definition of who is supposed to do what by which date (make it easy for them!)

| We communicated often via Skype to clarify issues and tasks

| keep in touch with them (but not by email)

| We signed Letters of Intent as part of the proposal

| demonstrates commitment

| makes management aware of the project


What I believe made KIWI a success (Formalia)

| We made use of the pre-proposal check service offered by

the Commission, which gave valuable feedback

| We tried to take into account all available background

material:

| work programme and guide for proposers

| background material specific to the unit/strategic objective

| slides of the Call 1 Information Days

| general EU policy documents


Challenges

08.09.2008 KIWI Project Presentation © 2008, Sebastian Schaffert, Salzburg Research

Challenges KIWI won’t solve (but it would profit from solutions!)

| efficient, distributed content and knowledge repository

| with reasoning support (rule-based as well as other)

| with support for both content and metadata

| and with support for billions of information items (content,

triples, ...)

| KIWI will only provide a “small and specific” solution!

| non-textual content

| (semi-)automatic metadata extraction

| annotation

| breaking the “format lock”

| and of course lots of other topics that are outside the direct

project scope (e.g. other Semantic Social Software) ...

Conclusion

Concl

usi

on


Summary (1)

| KiWi is concerned with knowledge management following

the wiki philosophy and applying semantic web technologies

| KiWi is evaluated in two use cases, software knowledge

management (Sun Microsystems) and project knowledge

management (Logica)

| KiWi core concepts are content items, tags, triples, users,

roles, and context

| the KiWi user interface builds on flexible, customisable

layouts and widgets

| KiWi is implemented using legacy Java EE technology using

a model-view-controller paradigm

Concl

usi

on


Summary (2)

| KiWi also addresses novel research issues in the areas of

reasoning, reason maintenance, information extraction and

personalisation

Concl

usi

on
