OntologyBased Knowledge Representation for a DomainIndependent ProblemSolving ITS Framework

1
Ontology-Based Knowledge Representation for a
Domain-Independent Problem-Solving ITS Framework
SWEL'08 _at_ ITS'08 Ontologies and Semantic Web
for Intelligent Educational Systems In
conjunction with the 9th International
Conference on Intelligent Tutoring Systems
Montréal, Canada, June 23-27, 2008

• Jean-François Lebeau, Mikaël Fortin,
• Amir Abdessemed and André Mayers

2
Overview

• Introduction
• ASTUS Framework
• Ontology-based KR Approach
• DL-based representation
• Advantages and limitations
• Conclusion

3
Introduction

• Goal Show an approach to use ontologies to
  facilitate the modeling of knowledge in ITS

• What kind of ITS?
• Problem-Solving ITS
• ITS based on a cognitive model
• ITS following the behavior described in the KVL
  Tutoring Framework VanLehn06
• ITS for well-defined domains
• Domain-independent ITS

4
KVL-Framework

• The Task domain is defined with knowledge
  components
• Outer loop (task/problem level) and Inner loop
  (step level)
• A Step corresponds to an action in the learning
  environment UI
• Inferences (learning events) correspond to the
  mental application of a knowledge component
• Steps follows one or more inferences

5
Cognitive models

• Different knowledge components types for
  procedural and declarative knowledge
• Models built with a production system
• Procedural knowledge is modeled with production
  rules
• Declarative knowledge is modeled with facts
• Procedural knowledge compiled declarative
  knowledge Anderson95

6
ASTUS Framework (principles)

• Domains modeled with a pedagogical POV
• All the knowledge components must
• Have a single, precise meaning
• Be modeled with glass-box formalisms when they
  are of pedagogical interest and efficient
  black-box formalisms when they are not
• Use the formalisms that facilitate their
  interpretation by the higher-level processes
• Declarative knowledge has a support role

7
ASTUS Framework (system)
8
ASTUS Framework (EA/LA)

• Expert Agent
• Step generation and plan recognition
• Simulations to verify the model and evaluate the
  problems
• Learner Model Agent
• Deduction of the applied knowledge component
  (when EA faces ambiguities)
• Assessment of the learners mastery of the
  knowledge components

9
ASTUS Framework (PA/IA)

• Pedagogical Agent
• Generation of feedback according to the outputs
  of EA and LA (inner loop)
• Selection of the next task (outer loop)
• Interfacing Agent Fortin08
• Production of the feedback on the UI
• Step recognition
• Communication of the knowledge components to the
  learner as UI elements

10
Knowledge in ASTUS

• Declarative knowledge is divided in
• Semantic (factual) knowledge
• Episodic (autobiographical) knowledge
• Goals (intentions, not state)
• Procedural knowledge is divided in
• Complex procedures (mental plan)
• Primitive procedures (step)
• Rules Queries (perception and mental op.)
• -gtCPs, RQ are inferences

11
Semantic components

• Chunks (DL-Concepts, OWL-Classes)
• Concepts (building blocks for problem solution)
• Relations (weak, optional or temporary
  relationship)
• Functions (unique relationship)
• Contexts (one for each stage of the task)
• Attributes (DL-Roles, OWL-Properties)
• Used for defining features
• Link a chunk instance to data or to another
  instance
• Unknown value
• Sharable among different chunks

12
Instances

• Instances (OWL/DL-Individuals)
• Usually specific for a task
• Static instances at the chunk level
• Classification rules
• Add Is-a relationships to an instance
• Instantiation rules
• Create the relations verified in the KB
• Find the image of a function or create it
• Set of instances (1-gtN attributes)

13
Fraction lab (Semantic)
IA creates a instance from the learners
input (24)
14
Fraction lab (Procedural)
15
DL-Based representation

• Development-time parallel representation using
  the Web Ontology Language (OWL-DL variant)
• Formal definition of the semantic components
• Defined classes (set of conditions, restrictions,
  expressions, )
• Disjoint classes
• Properties characteristics (inverse, transitive,
  )
• Why is that useful ?
• Avoid simple/multiple inheritance issues
• Detect inconsistencies
• Discover hidden taxonomy relationships
• Pellet is used for reasoning
• Protégé-OWL is used for visualization

16
Advantages

• Handling instances added to the WM
• A method to handle learners input
• A method to reduce domain-specific efforts
• Clear distinction between different semantic
  knowledge components to facilitate their
  interpretation by LA and PA
• Expert Systems rules backed up with DL-based
  ontology checking

17
Limitations

• No OWL representation at runtime
• Limited interpretation of semantic knowledge
• No DL-based reasoning
• Could use the TBox at runtime
• No glass-box formalisms for rules
• Some help still possible if the skill is not
  mastered
• SWRL wouldnt do much better than Jess

18
Conclusion

• What the ASTUS aims to offers for developing
  Problem-Solving ITS
• Some advantages of an ontology-based
  representation for the semantic knowledge in this
  context
• An original way to use ontologies in
  Problem-Solving ITS

19
Future work

• Link different labs with points of contact in
  their ontologies (and a shared goal)
• Develop the top-level ontology to enable the use
  of problem-solving methods
• More complex semantic knowledge components
• A way to tackle less well-defined domains

20
Questions ?Also, come to see us at the demo
session later today !
21
References

• Anderson, J. R., Corbett, A. T., Koedinger, K. R.
  et Pelletier, R., 1995. Cognitive tutors
  Lessons learned. Journal of Learning Science 4
  (2), 167-207.
• Fortin, M., Lebeau, J-F. Abdessemed A.,
  Courtemanche, F. and Mayers, A. A Standard Method
  of Developping User Interfaces for a Generic ITS.
  The 9th International Conference on Intelligent
  Tutoring Systems (ITS 2008). June 23-27,
  Montréal, Qc, Canada.
• VanLehn, K. (2006) The behavior of tutoring
  systems. International Journal of Artificial
  Intelligence in Education. 16.
• Jess http//www.jessrules.com/
• Pellet http//pellet.owldl.com/
• Protégé-OWL http//protege.stanford.edu/

22
Cut slides follow
23
Contexts

• Domain-independent contexts
• Ontological (static instances)
• Reasoning (rules-created and inputs)
• Knowledge Base (context instances)
• Domain specific contexts
• Used by IA as a Model (MVC) for each stage of the
  task where the UI is different
• -gt Complete hierarchy from KB to instances

24
Knowledge Base

• Read access trough queries
• Queries return snapshots of instances
• Only attributes values are saved
• Episodic knowledge components
• Write access trough scripts
• Trigger primitive procedures, context
  initialization and context transition
• Effects Add/Modify/Remove instances
• Add always done through a context instance

25
Simulation kernel

• Domain-specific objects grafted to instances
• Scripts use them to produce their effects
• Useful to generate instances from calculation
• Genetics lab
• Can also have an internal state
• Requirement for simulator-based labs

26
Procedural vs Declarative

• Declarative knowledge
• Perimeter of a triangle def. P a b c
• A representation of a triangle Ta, b, c
• An concrete triangle T1(a3, b4, 55)
• Procedural knowledge
• A skill that creates P1 from T1
• Declarative knowledge in a compiled form
  Anderson95

27
(No Transcript)
28

• Back