Transcript Exploiting Semantic Web and Knowledge Management

Exploiting Semantic Web and
Knowledge Management
Technologies for E-learning
Sylvain Dehors
Director
Rose Dieng-Kuntz
INRIA Sophia Antipolis
University of Nice-Sophia Antipolis/ ED STIC
E-learning, this ?
2
A vision of e-learning
• For us:
– Any learning activity mediated by a computer
– Buzz Word, but also real change in practices
• Use of computers in daily activities
• All ages, from youngster to adult teaching
• In practice, several types of application
– Simulation programs
– Tutoring systems
– On-line courses
3
Our e-learning situation
• Learning organization
– Teacher(s) with a group of students
• Environment
– Computers for daily usage
– Either on-line or face-to-face
• Knowledge Sources
– Course documents
– Teacher’s expertise
Provide computer support for taking
advantage of the knowledge sources
4
Outline
1. Research question
2. Method Proposal
1. Selection and analysis of existing material
2. Semi automatic annotation
3. Learning activity
4. Analysis
3. Conclusion
5
Research question
How can teachers and students better use
knowledge sources, such as pedagogical
documents, with computer interfaces ?
• Proposal:
– apply Knowledge Management techniques
and Semantic Web technology
– develop a practical method
• Illustration: a tool (QBLS) and experiments
6
Inspirations
• Knowledge Management
– “The objective of a knowledge management structure
is to promote knowledge growth, promote knowledge
communication, and in general preserve knowledge
within the organisation” (Steels L., 93)
• Semantic Web:
– “The Semantic Web provides a common framework
that allows data to be shared across application,
enterprise, and common boundaries.” (W3C)
– Standards: RDF, RDFS, OWL, SPARQL
7
Existing methods and tools (Dieng et al.)
• Corporate semantic web
Knowledge
holder
DB
ontologies
edit A
edit O
Knowledge Management Syst.
User (collective task)
documents
services
Semantic
annotation
base
query
User (Individual task)
• Apply to a learning organization
- Tool: Corese semantic search engine to query formalized
knowledge
- W3C Standards expressing knowledge about the course
8
Method description
1 - Selection and
analysis of
existing
material
1
4
4 - Analysis
2
3
2 - Semi automatic
annotation
3 - Learning activity
9
Select
Original
resources
selection
Method description
1
KM tools
2
Enrich
Semantization
Ontologies :
Document
Pedagogy
Domain
Usage
feedback
tests
Annotations
Activity
analysis
4
Analyze
3
Use
Conceptual
navigation
+ adaptation
10
Experiment’s Agenda
QBLS-1 :
QBLS-2 :
QBLS-ASPL :
2 hours lab
3 months course
Knowledge Web NoE
Signal Analysis
Java Programming
Semantic Web studies
2005
2006
2007
11
Resource selection
• First, establish a pedagogical strategy
– Collaboration Teacher/QBLS designer
– QBLS: Question Based Learning Strategy: Motivation,
autonomy, self-directed learning
• Existing resources:
– Objective criteria
• Availability, standard editable format (XML)
• Suitability for annotation (modularity, coherence, vocabulary
used)
– Subjective criteria
• Scope, goal, context
• Teacher’s acceptance
12
Original documents
Power Point presentations
– Signal analysis / Java programming
– Used as hard copy course material
Modularity
Coherence, Vocabulary
13
Ontology selection
• Selection of existing models, ontologies?
– Document:
• Must fit the course structure
• Document organization
Document ontology
– Pedagogy:
• Appropriate for the pedagogical approach
– Domain to learn:
• Usually the biggest ontology
• Fit the document contents (vocabulary used, conceptualization)
• Fit the teacher’s vision
Lots of constraints, difficult to find appropriate ontologies
14
1- Selection and
analysis of
existing material
4 - Analysis
2 – Semi automatic
annotation
3 – Learning activity
15
Annotation
• Express additional knowledge about the
course
– Based on teacher’s expertise and vision
• Principles :
– Use existing edition tools
– Proceed through visual mark-up
– Rely on XML technologies and Semantic Web
formalisms
16
A semi-automatic process
• 3 steps
– Pre-processing
– Manual annotation
– Automatic extraction
resource to
reuse (XML)
“annotable”
version
annotated
version
content
(XHTML)
annotation
pre-processing
manual
annotation
xsl
transform.
(RDF)
Ontologies
(OWL, RDFS)
17
Preprocessing
• Identification of the content characteristics
– Separation in small entities
• Automatic annotation
– Vocabulary used → domain concepts,
automatic annotation with domain ontology
– Resource roles → pedagogical ontology
• Preparation
– Styles → reflect ontological concepts
– enrich style lists with ontologies
18
Preprocessing
22
Manual annotation
• Exploitation of tools functionalities by the teacher
for a visual markup
• Evolution/enrichment/creation of corresponding
domain ontology
• Practical objective: connecting navigation paths
– Edition of the content
– Linking concepts with semantic hierarchical relations
(SKOS)
Interface
skos:broader
Keyword « implements »
skos:broader
Conditional
Statement
Statement
skos:broader
Assignment
Statement
23
Final result: Open Office-Writer
24
Final result : MS-Word
25
Experimental results: ontology re-use
• Pedagogical ontology
– Reused directly
– Same intention as original: describe ped. role
(generic?)
• Domain Ontology
– Design intention very important: here offer
“conceptual views” of the resources
– Mostly developed specifically, comparisons with other
domain ontologies show striking incompatibilities.
Method modifiers
Access rights
public
protected
private
public
protected
private
26
Experimental results: annotation cost
Number of resources
QBLS-1
QBLS-2
92
359
Num. of resources discarded None
54
Course duration
2H
3 months
Number of pedagogical
types used (directly)
8/8
12/27
Num. of domain concepts
41
171
Editing Tool
Microsoft Word
OpenOffice Writer
Annotation time
N/A
20H
Modification of content
Yes
No
27
1- Selection and
analysis of
existing material
4 - Analysis
2 – Semi automatic
annotation
3 – learning activity
28
Learning activity
• Offer “conceptual” navigation in the set of
resources while answering questions or
performing exercises
• Navigation through semantic queries
– Take advantage of domain concepts hierarchy
(broader links)
– Use typology of pedagogical concepts for ordering
(subsumption)
• Interface generation
– Static XSL style sheets: performance, reuse,
maintenance
29
Semantic Web architecture
Domain
vocabulary
(Skos)
Doc. model
Corese
Semantic
Search Engine
(RDFS)
rules
Pedagogical
ontology
(OWL)
logs
(RDF)
4
3
Answers
(Sparql-XML)
Queries
(Sparql)
Formalized Knowledge
web-app
content
(XHTML)
XSLT
Learner
2
Request
Interface
(XHTML)
Tomcat web server
5
6
HTTP
1
30
Semantic Web at work
• Dynamic SPARQL queries:
Variable
skos:primarySubject
SELECT * WHERE {
FILTER (?c = java:variable)
{ ?doc skos:primarySubject ?c }
UNION
{ ?doc skos:primarySubject ?c2 . ?c2 skos:broader ?c}
Local
Variable
skos:primarySubject
rdf:type
?doc rdf:type ?t
?t edu:order ?order
?doc dc:title ?docTitle
?t rdfs:label ?docLab
?c skos:prefLabel ?cLab
}
ORDER BY ?order
skos:broader
Definition
rdf:type
edu:order
Layout
information
3
Example
edu:order
7
31
QBLS-1
33
QBLS-2
Human readable information
Variable
Fields
Local
variable
34
Experimental results: students’ feedback
QBLS-1
QBLS-2
Num. of students using the system
100%
30%
Num. of resources visited
90%
80%
Overall Satisfaction
4.3/5
3.9/5
Off-hours access
N/A
50% of connections
• Good satisfaction
• Structured navigation appreciated for
direct access to information
• Use of domain and pedagogical
information
35
QBLS-ASPL
(Advanced Semantic Platform for Learning)
• Existing resources on a portal : REASE,
• MS-PowerPoint files
36
QBLS-ASPL
Interesting
Web sites
for
advanced
learners
37
QBLS-ASPL
Provided by
QBLS
38
1- Selection and
analysis of
existing material
4 - Analysis
2 – Semi automatic
annotation
3 – Exploitation by
learners
39
Analysis
• Modeling user activity
– A navigation model based on a graph representation
Concept
User A
subject of
Resource
mentions
Concept
subject of
Resource
Time t
• Exploitation of logs
– Visualization through automatically generated graphs
– Use semantic querying to highlight particular
characteristics of the graphs represented in RDF
40
Visualization
41
Visualization
42
Semantic querying
• Find regularities, patterns?
– Using the graph structure
– Relying on the ontology
SELECT ?user count ?v WHERE {
?aux skos:primarySubject ?concept
?aux rdf:type edu:Auxilliary
?v edu:user ?user
?v edu:conceptVisited ?concept
OPTIONAL { ?v2 edu:resourceVisited ?aux
?v2 edu:user ?user}
FILTER(! bound(?v2))
}
?v
Object
Def
?v2
Ex.
43
Experimental Results
• Involve teacher’s in the analysis
– Problem with large size graphs
– Visualization tools not sufficient yet
– Needs to be coupled with other sources of
information
• First step towards automated interpretation
– Define a collection of patterns -> behavioral
patterns
• Use in “real-time”?
44
Conclusion
Learning Object Repositories
LOM standard
Annotation tools
Linguistic analysis
Scorm?
Learner modeling
Activity tracking
Learning Design
Adaptive hypermedia
Semantic Web = valid connector
45
Conclusion (2)
• Semantic web interests:
– Existing tools, Corese, Protégé, etc.
– Existing models, in standard language
– Unification and connection with other systems
• Ontologies for e-learning
– Interest, reusability of domain might be limited
– Need for simple expressivity, “goal oriented
design”
46
Conclusion(3)
• Resource Reuse
– Observed use and good satisfaction level
– Definite interest, cost still high
• Knowledge management approach
– Satisfaction of users
– Initial goal fulfilled
– May apply to other learning contexts
47
Perspectives (1)
• Short term
– Further develop annotation system based on existing
tools
– Administrative tools to make teachers fully
autonomous
• Middle term
– Enhance scalability with large RDF bases ( when
triples are generated by learner activity)
– Generalize log visualization, work on usage of such
representations (e.g. teachers’ interpretations)
48
Perspectives (2)
• Long term
– Investigate the cognitive implications for
learning of the annotations
• Importance of the pedagogical concepts
• Structure of the domain
– Enhance user tracking (more information,
refine model)
49
Acknowledgements
• Catherine Faron-Zucker
• Jean Paul Stromboni
• Peter Sander
50