This document discusses using semantic techniques to enable knowledge reuse in conceptual modeling. It describes how qualitative reasoning models can be semantically grounded and aligned using ontologies to provide ontology-based feedback. The techniques allow relating learner models to expert models to automate acquiring feedback and recommendations. An evaluation shows semantic web resources can ground domain vocabularies and ontology matching tools can align conceptual models. The approach supports creating semantically networked models to share and reuse conceptual knowledge.
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
1. Semantic Techniques for Enabling
Knowledge Reuse in Conceptual
Modelling
Jorge Gracia, Jochem Liem, Esther Lozano, Oscar Corcho,
Michal Trna, Asunción Gómez-Pérez, and Bert Bredeweg
Ontology Engineering Group, Universidad Politécnica de Madrid. Spain
{jgracia, elozano, ocorcho, mtrna, asun}@fi.upm.es
Informatics Institute. University of Amsterdam. The Netherlands
{j.liem, b.bredeweg}@uva.nl
ISWC, November 10th, Shanghai, China
2. Outline
2
1. Introduction
1. Qualitative Reasoning Modelling and Simulation
2. Application of Semantic Techniques
3. Existing approaches
2. Semantic Grounding
3. Ontology-Based Feedback
1. Types of feedback
2. Techniques of OBF
4. Some numbers
5. Conclusions
6. What’s going on...
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
3. Introduction
• Conceptual representation of physical
systems
• Prediction of the system behaviour
through reasoning
• Simulation
• Qualitative (important landmarks no
numerical details)
• Separation of structure and behaviour
• Multiple domains of application
• Environmental science
• Physics
• Economy
• ...
3
Qualitative Reasoning
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
4. Introduction
• Learning by modelling approach
• Learners formally express and test their
conceptual knowledge about systems
in an educational context
• Desirable features:
– Shared learning environment, where
expert and learner models are uploaded
– Quality feedback extracted from the
common knowledge
4
QR in science and education
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
5. QR Modelling and Simulation
5
Knowledge Representation
Entity hierarchy Scenario
Quantity:
The dynamic aspects of
the system
Influence:
Natality determines δSize
Proportionality:
δSize determines δNatality
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
6. Semantic Techniques
• Bridging the gap between the loosely and imprecise
terminology used by a learner and the well-defined semantics
of an ontology
• Relating the QR models created by other learners or experts in
order to automate the acquisition of feedback and
recommendations
6
How semantic techniques could help?
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
9. Existing approaches
• QR modelling and simulation tools: Betty’s brain, Stella
• Not grounding of terms to a common vocabulary
• Not quality feedback from other models
• Conceptual modelling techniques: CmapTools
• Concepts maps for knowledge representation
• Collaborative use
• No use of Semantic Web to maximize interoperability
• Not common shared vocabularies
• Semantic techniques to enhance collaborative learning:
DEPTH (Design Patterns Teaching Help System)
• Focused on software engineering education
• Supports recommendation more rather than quality feedback
9
Related work
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
10. Outline
10
1. Introduction
1. Qualitative Reasoning Modelling and Simulation
2. Application of Semantic Techniques
3. Existing approaches
2. Semantic Grounding
3. Ontology-Based Feedback
1. Types of feedback
2. Techniques of OBF
4. Some numbers
5. Conclusions
6. What’s going on...
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
12. Semantic Grounding
• Ensure lexical and semantic correctness of terms
• Ensure the interoperability among models
• Extraction of a common domain knowledge
• Detection of inconsistencies and contradictions between
models
• Inference of new, non declared, knowledge
• Assist the model construction with feedback and
recommendations
12
Benefits of grounding
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
13. Outline
13
1. Introduction
1. Qualitative Reasoning Modelling and Simulation
2. Application of Semantic Techniques
3. Existing approaches
2. Semantic Grounding
3. Ontology-Based Feedback
1. Types of feedback
2. Techniques of OBF
4. Some numbers
5. Conclusions
6. What’s going on...
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
14. Types of OBF
• Ontology matching based feedback
• Improvements of terminology
• Missing and extra terms in the learner model
− Missing hierarchical relations
• Semantic reasoning based feedback:
Inconsistency between hierarchies of models
• Structure comparison based feedback:
Differences between model structures
14Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
15. Modules of OBF
15
Ontology
Matching
Semantic
Reasoner
Missing/Extra
Ontology Elements
List of
differences
Structure
Comparison
Grounding-Based
Alignment
Learner Model
+
Reference Model
Inconsistency
between Hierarchies
Differences between
Model Structures
Improvement of
Terminology
Preliminary
mappings
List of equivalent
terms
17. Ontology Matching
• Ontology matching tool: CIDER
• Input of the ontology matching tool
• Learner model with preliminary mappings
• Reference model
• Output: set of mappings (Alignment API format)
17
J. Gracia and E. Mena. Ontology matching with CIDER: Evaluation report for the OAEI 2008. In Proc. of 3rd Ontology Matching Workshop
(OM’08), at ISWC’08, Karlsruhe, Germany, volume 431, pages 140-146. CEUR-WS, October 2008
18. Ontology Matching
18
Improvement of terminology
Learner model: Reference model:
equivalent terms with different label
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
19. Ontology Matching
19
Missing and extra terms in the learner model
Learner model: Reference model:
missing term
extra term
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
20. Ontology Matching
20
Missing hierarchical relationships
Reference model:
missing term
equivalent terms
Learner model:
subclass of
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
21. Semantic Reasoning
21
Inconsistency between hierarchies
Reference model:
Learner model:
equivalent terms
Disjoint classes
INCONSISTENT
HIERARCHIES!
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
22. Structure Comparison
22
Missing QR structures
Learner model: Reference model:
missing model structures
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
23. Outline
23
1. Introduction
1. Qualitative Reasoning Modelling and Simulation
2. Application of Semantic Techniques
3. Existing approaches
2. Semantic Grounding
3. Ontology-Based Feedback
1. Types of feedback
2. Techniques of OBF
4. Some numbers
5. Conclusions
6. What’s going on...
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
24. Experimental Evaluation
Q1. Are Semantic Web resources suitable for
grounding specific domain vocabularies?
Q2. Are the state-of-the-art ontology matching tools
suitable for the alignment of QR models?
24
Some Research Questions
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
25. Q1 – Semantic Web resources
Knowledge source Coverage Ratio
DBpedia 72%
OpenCyc 69%
WordNet 45%
Watson 47%
25
Tested 1686 different English words coming from DynaLearn glossaries
Knowledge source Coverage Ratio
DBpedia + Yahoo Spelling Suggestion 78%
What if we fix spelling errors and suggest nearby terms?: “fiter feeding” “filter feeding”
Coverage Study
What if we combine several sources?:
Knowledge source Coverage Ratio
DBpedia + OpenCyc 87%
DBpedia + Watson 73%
Dbpedia + WordNet 72%
Dbpedia + OpenCyc + WordNet + Watson 88%
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
26. Q1 – Semantic Web resources
• Tested 909 English labels covered by DBpedia, randomly selected from
DynaLearn glossaries
• Asked 8 expert evaluators, each one evaluated between 200-300
groundings. Each grounding was double-evaluated
• Question: For each grounded term, are all the relevant meanings that
you have in mind contained in the set of grounding candidates? If yes,
mark the relevant ones.
• Average accuracy: 83%
26
Accuracy Study
Inter-evaluator Agreement Level
Polysemy Agr eement (any) Cohen's Kappa
21,35 85% 0,47
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
27. Q2 - Ontology Matching
• Golden standard defined by experts
• Eight QR models grouped by pairs
• Semantic equivalences between them were identified
• Result: reference alignment file
• Separated ontology alignment: CIDER, Falcon
• Each produced alignment was compared to the golden standard
27
Model Matching Experiment
Precision Recall
CIDER 92% 95%
Falcon 67% 95%
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
W. Hu and Y. Qu. Falcon-ao: A practical ontology matching system. Journal of Web Semantics, 6(3):237-239, 2008
28. Outline
1. Introduction
1. Qualitative Reasoning Modelling and Simulation
2. Application of Semantic Techniques
3. Existing approaches
2. Semantic Grounding
3. Ontology-Based Feedback
1. Types of Feedback
2. Techniques of OBF
4. Some numbers
5. Conclusions
6. What’s going on...
28Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
29. Conclusions
• Support the creation of semantically networked models to
share and reuse conceptual knowledge
• QR models are exported into an ontological language and
grounded to an external common vocabulary
• Ontology matching techniques used to get quality feedback
29
Conclusions
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
30. Outline
1. Introduction
1. Qualitative Reasoning Modelling and Simulation
2. Application of Semantic Techniques
3. Existing approaches
2. Semantic Grounding
3. Ontology-Based Feedback
1. Types of feedback
2. Techniques of OBF
4. Some numbers
5. Conclusions
6. What’s going on...
30Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
31. What’s going on…
31
Future Work
• Run usability studies on our ontology-based feedback
• Enrich our ontology matching based techniques with
advanced metrics
• Provide OBF results through the virtual characters
• Model recommendation based on the community of users
• Use our system in the academic domain to support
semantic-guided learning
• Publish this semantic data in the web of Linked Data
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
32. What’s going on…
32
OBF through virtual characters
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
Differences in terminology 2/8
The quantity Death rate of your model is
equivalent to the quantity Mortality of the
reference model. More info.
Do you want to rename the term as Mortality?
Yes No
33. Thanks for your attention!
33
Jorge Gracia
Facultad de Informática
Universidad Politécnica de Madrid
Campus de Montegancedo sn
28660 Boadilla del Monte, Madrid
http://www.oeg-upm.net
jgracia@delicias.dia.fi.upm.es
Phone: 34.91.3363670
Fax: 34.91.3524819
Some images under Creative Commons licence :
http://www.flickr.com/photos/binkley27/2969227096/
http://www.flickr.com/photos/tauntingpanda/14782257/
http://www.flickr.com/photos/rainforest_harley/232636845/
Semantic Techniques for Enabling Knowledge Reuse in Conceptual Modelling
Notas do Editor
QR models are conceptual model. Purpose is communication
Conceptual models have an inherent ontology (particular perspective on the world)
QR models are domain independent (all domains that can be described by system dynamics)
Allows both scientific theories to be captured, but also thoughts of students
QR models allow simulations. However they are qualitative in nature
Qualitative means important values are made explicit Heavy metal concentration = {zero, positive, legal limit, illegal}. Its an abstraction from numerical values.
QR models use a compositional modelling paradigm. That is, each distinct model part is represented as a model fragment.
There is a strict separation of structure and behaviour. The structure describes the components of the system. The behaviour describes the dynamic aspects of the system. This separation allows the correct model fragments to be found that apply to a particular system.
A simulation is based on an initial situation (scenario). The result of a simulation is a state graph. Each state represents a possible situation the system can be in. The transitions indicate how the system can change from one situation to another.
Within DynaLearn domain experts are creating models to support an environmental science curriculum.
There will be 100s of expert models in a semantic repository. Learners will also upload their models to the repository. We want to use this resource. Grounding makes this possible.
Model = QR model as a type of conceptual model (not the same as a model in logics…)
- Semantic Repository to storage the models and domain vocabularies
- Grounding of terms to the common vocabulary
- Ontology-based feedback on the quality of the models
- Recommendation of models and model fragments based on model features or collaborative filtering