Domain Modeling for Personalized Learning
Peter Brusilovsky School of Information Sciences,
University of Pittsburgh
What is the Domain Model?
• What you are using it for? • A personalized learning prospect: sequencing, navigation
support, and recommendation research • Enumeration of domain knowledge • Serve as a basis for individual student models • Serve as a way to describe, classify and index learning
content • Provide connections between state of learner knowledge
and relevant content • to model the learner after interaction with content
(question, step, example, chapter…) • to decide what is the next best thing to learn
Why Do We Need Domain Models?
• Following Sleeman – Sleeman, D.H.: UMFE: a user modeling front end system.
International Journal on the Man-Machine Studies 23 (1985) 71-88
• User models can be classified by the nature and form of information contained in the model as well as the methods of working with it
– Brusilovsky, P. and Millán, E. (2007) User models for adaptive hypermedia and
adaptive educational systems. The Adaptive Web: Methods and Strategies of
Web Personalization, Springer-Verlag, pp. 3-53.
Classifying Domain Knowledge Models
Three “Sleeman” Layers
• Nature – what is being modeled
• Structure – how this information is represented
• Functionality – how models are used
• Tools – how we (ITS experts) can work with it
Structured Doman Models
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N
• AKA Network of “Things”
• Most of the models can be represented in this form
• What is the nature of each DM element?
• How these elements are organized?
Nature: Kind of Knowledge
• What kind of knowledge DE represents? • Procedural (interpretable)
– How things work? (simulation) – How to construct things? (building) – How to evaluate results? (i.e., constraints)
• Conceptual (representational) – What do you know?
Nature: Granularity of Elements
• What is the granularity of modeling? • Procedural
– Rules – Procedures and plans
• Conceptual – Facts – elementary units, 1000s for a domain (AI experts) – Concepts – fine grain, 100s for a domain (domain experts) – Topics – coarse grain, 10s for a domain (teachers)
• Only low level KEs can be considered “cognitive” and checked with curves
Structure
• Vector Models (Enumerative) • Network models (Structured)
– Clusters – Hierarchy with single connection type – Heterarchy or network with multiple
connection types
Vector Model of Knowledge
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N
No connections, just enumeration
Network Model of Knowledge
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N
Connections represent additional knowledge, help in modeling and adaptation
Classic Bug Model
Rule A Rule B
Rule C
n Classic Bug Model is formed by independent rules (skills) with each having various malrules (misconceptions)
More Advanced Network Procedural Models
• Pedagogical links (prerequisites) • Skill Hierarchy
– Procedure -> Steps - > Substeps – GOMS
• Genetic Model – Adds genetic relationships that represent the
advancement of skills on different levels of mastery – Goldstein, I. P. (1979) The Genetic graph: a representation for the evolutionof procedural
knowledge. International Journal on the Man-Machine Studies 11 (1), 51-77.
Conceptual Models
• Almost all finer-grain conceptual models are network models
• Semantic Models on the level of facts – Buenos Aires is a capital of Argentina
• Classification hierarchies (is-a) • Decomposition hierarchies (part-of)
Decomposition Model in ADAPTS
• Hierarchy of Domain objects – System/Subsystem – Replaceable Unit – Addressable Unit
• Different levels of components correspond to different kinds of knowledge the user may have
Aircraft (SH-60)
Sonar
Subsystem 1 Subsystem 2
Subsystem 1.2Subsystem 1.1
Replaceable Unit A Replaceable Unit B
. . .
. . .
Addressable Unit X Addressable Unit Y
. . .
Brusilovsky, P. and Cooper, D. W. (2002) Domain, Task, and User Models for an Adaptive Hypermedia Performance Support System. In: Y. Gil and D. B. Leake (eds.) Proceedings of 2002 International Conference on Intelligent User Interfaces, San Francisco, CA, January 13-16, 2002, ACM Press, pp. 23-30.
Classification Model: Tree of Life
• Tree of Li
Conceptual Modeling with Ontologies
• Modern approach to domain modeling used ontological frameworks
• Allows to represent multiple types of connections
• Many standard tools and approaches to use from Semantic Web (development, extraction…)
• We use ontologies for the last 10 years for all domain modeling work
Ontologies for Domain Modeling
• Created ontologies for C, Java, SQL domains • Ontology-based content indexing
– Hosseini, R. and Brusilovsky, P. (2013) JavaParser: A Fine-Grain Concept Indexing Tool for Java Problems. In: Proceedings of The First Workshop on AI-supported Education for Computer Science (AIEDCS) at the 16th Annual Conference on Artificial Intelligence in Education, AIED 2013, Memphis, TN, USA, July 13, 2013, pp. 60-63, also available at https://sites.google.com/site/aiedcs2013/proceedings.
• Ontology mapping for multi-system personalization – I.e, Database Exploratorium and Mitrovic SQL Tutor – Sosnovsky, S., Brusilovsky, P., Yudelson, M., Mitrovic, A., Mathews, M., and Kumar, A. (2009)
Semantic Integration of Adaptive Educational Systems. In: T. Kuflik, S. Berkovsky, F. Carmagnola, D. Heckmann and A. Krüger (eds.): Advances in Ubiquitous User Modelling. Lecture Notes in Computer Science, Vol. 5830, pp. 134-158.
Ontological Domain Model for Java • Java Ontology
specifies about 500 classes connected with 3 types of relations: subClassOf, partOf/hasPart, and related
• About 300 classes are available for indexing
• A class can play one of two roles in the problem index: prerequisite or outcome
[20]
Aspect-based Conceptual Modeling in ADAPTS
CONCEPT Reeling Machine
CONCEPT Sonar Data Computer
CONCEPT Sonar System
Removal Instructions
Testing Instructions
Illustrated Parts
Breakdown Principles
of Operation
Principles of
Operation
Principles of
Operation Removal
Instructions
Removal Instructions
Testing Instructions
Testing Instructions
Illustrated Parts
Breakdown
Illustrated Parts
Breakdown
[21]
User model: multiple aspects, multiple evidence
Certified
CONCEPT Reeling Machine
CONCEPT Sonar Data Computer
CONCEPT Sonar System
ROLE Removal
Instructions
ROLE Testing
Instructions
ROLE IPB
Reviewed Hands-on
Simulation
AT2 Smith
AD2 Jones
Preference
Reviewed
Hands-on +
Certified
Reviewed
Hands-on
Hands-on Reviewed
Reviewed
ROLE Theory of Operation
Application of Domain Models
• Basis for overlay student models • Basis for content indexing (i.e., which problem,
example, step, page fragment related to which KE?
• Taken together, it enables – Student Modeling an Open Student Modeling – All kinds of personalized guidance (i.e., when to stop,
what is next…) – All kinds of adaptive presentation
Simple overlay model
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N yes no
no
no yes
yes
Simple overlay model
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N yes no
no
no yes
yes
Weighted overlay model
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N 10 3
0
2 7
4
Topic-based Content Indexing
Example 2 Example M
Example 1 Problem m
Example N Problem K
Topic 1 Topic 2
Topic N
Problem 1
Problem 2
Problem 10
Each content item is assigned to one topic
Concept-based Content Indexing
Example 2 Example M
Example 1
Problem 1
Problem 2 Problem K
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N
Examples
Problems
Concepts
Each content item is indexed with several concepts Brusilovsky, P. (2003) Developing Adaptive Educational Hypermedia Systems: From Design Models to Authoring Tools. In: T. Murray, S. Blessing and S. Ainsworth (eds.): Authoring Tools for Advanced Technology Learning Environments: Toward cost-effective adaptive, interactive, and intelligent educational software. Kluwer: Dordrecht, pp. 377-409.
Personalized Guidance
• When to stop? Typical use of skill models – Mastery learning
• What to do next? Typical use of concept models • Which knowledge to learn? Knowledge sequencing • How to learn it? Content sequencing
• Content sequencing (AI makes decision) – Questions, problems, examples, readings… – Proactive or remedial content sequencing
• Adaptive navigation support (Human + AI) Brusilovsky, P. (2007) Adaptive navigation support. In: P. Brusilovsky, A. Kobsa and W. Neidl (eds.): The Adaptive Web: Methods and Strategies of Web Personalization. Lecture Notes in Computer Science, Vol. 4321, Springer-Verlag, pp. 263-290.
QuizGuide: Topic-Based Nav. Support
Sosnovsky, S. and Brusilovsky, P. (2015) Evaluation of Topic-based Adaptation and Student Modeling in QuizGuide. User Modeling and User-Adapted Interaction 25 (4), In Press.
NavEx: Concept-based Navigation Support
Yudelson, M. and Brusilovsky, P. (2005) NavEx: Providing Navigation Support for Adaptive Browsing of Annotated Code Examples. In: Proceedings of 12th International Conference on Artificial Intelligence in Education, AI-Ed'2005, Amsterdam, the Netherlands, July 18-22, 2005, IOS Press, pp. 710-717
Mastery Grids Sequencing Service
Hosseini, R., Hsiao, I.-H., Guerra, J., and Brusilovsky, P. (2015) What Should I Do Next? Adaptive Sequencing in the Context of Open Social Student Modeling. In: Proceedings of 10th European Conference on Technology Enhanced Learning (EC-TEL 2015), Toledo, Spain, pp. In Press.
Indexing of Content Fragments
Fragment 1
Fragment 2
Fragment K
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N
Node"Concepts"
Adaptive Presentation in ADAPTS
Domain Modeling: How?
• Manual domain modeling – Knowledge Engineering – Expensive, needs several kinds of experts – Many authoring support systems (i.e., InterBook)
• Automatic, from text – Fact extraction – Rule and casual relationship extraction – Concept and link extraction (uni- bi- tri- grams) – Topic modeling (LSA, LDA) – Remedial content sequencing
Indexing: How?
• Manual domain modeling – Manual indexing by experts
• Powerful, expensive • Supported by many good authoring systems
– Crowdsourced indexing – Automatic step indexing (model tracing) – Automatic content indexing (i.e., Java Parser)
• Automatic, from text or usage data – Naturally automatic indexing – Scalable but limited use (i.e., texts, sometimes
questions)