X Som Graduation Presentation

1. An Ontology-Based Data Integration System: Solving Semantic Inconsistencies Giorgio Orsi October 23, 2006 X-SOM (eXtensible Smart Ontology Mapper) Politecnico di Milano

2. Introduction

This work is part of the Context-ADDICT project.

Context-dependent information systems.

Heterogeneous and mobile data sources.

Need for integration of different schemata.

Ontology

Shared and formal specification of a

conceptualization of a given domain.

O = KB =

OWL-DL (Web Ontology Language [DL])

3. Why Ontologies?

Well formalized, XML-based format for the description and integration of data source schemata.

Exploit ontology semantic richness for schema integration:

an ontology as a unified representation to mediate access to data sources.

an ontology as a guide for terminological/structural conflict resolution in traditional schema integration

X-SOM:

(Semi-)automatic ontology mapper.

Semantic consistency check.

Need for high precision of the results.

4. Ontology Matching Vehicle SportCar Car Truck Motorized Motorless SW SportCar Wheel Vehicle Bicycle 1.0 0.5 0.7 1.0 0.3 1.0

5. Approaches

Ontology Mapping : Binding ontology's concepts with owl:equivalentClass and rdfs:subClassOf properties.

Ontology Merging : Creating new concepts by merging data sources' elements in new ones.

Ontology Articulation : Creating a set of rules , describing relationships among data sources' concepts.

Ontology Integration : Creating a unique integrated ontology by reuse of the elements of data sources' ontologies as-is.

6. Additional Requirements Issues What we need

Tools requiring massive user's interaction.

Learning from user's corrections.

Ontologies created by designers with different technical backgrounds.

Management of the mismatches arising from different views of the application domain.

Many techniques for ontology mapping (syntactic, structural, probabilistic,...).

Combining different techniques considering their reliability.

7. X-SOM's Components XML/OWL OWL Loader OWL dialect verifier Mapper Mapping Strategy WordNet String Metrics Walk ISLab HMatch OWL Input Ontologies Output Ontology Writer I/O Mapper Modules Module Interface Module Interface Module Interface Module Interface Modules Registry Average Functions Registry XML XML Config. Files Consistency Checker Manual mapper Module Interface Google API Module Interface ... Module Interface Neural Network Trainer Training Set Builder GUI

8. Aggregation: The Neural Network y=ax y=ax y=ax y=ax Jaro Module WordNet Module ISLab HMatch Another Module y=sigm(x) y=sigm(x) y=sigm(x) y=sigm(x)

Determines the weight of each matching technique.

Supports the learning phase to achieve automatic behavior.

9. Consistency Checking

Standard Consistency Check: Ensure the T-BOX satisfiability.

Semantic Consistency Check: Performs standard consistency check with the addition of semantic local checks:

Bowties and cycles.

Multiple mappings.

Hypothesis : Consistent source ontologies.

10. Inconsistencies Sympthoms O2:Y O2:Z O1:X

Multiple mappings:

Loss of concepts satisfiability:

Master O1:Student Bachelor Ph.D. Master Bachelor O2:Student unionOf unionOf DisjointWith DisjointWith DisjointWith equivalentClass equivalentClass equivalentClass equivalentClass

11. Semantic inconsistencies Reader Author Organization Person Author Person subclassOf subclassOf subclassOf subclassOf equivalentClass

12. Semantic inconsistencies Reader Author Organization Person Author Person subclassOf subclassOf subclassOf subclassOf equivalentClass Everything that is an author is also a person An organization is a person Wrong!

13. Semantic inconsistencies Reader Author Organization Person Author Person subclassOf subclassOf subclassOf subclassOf equivalentClass

14. Testing X-SOM

15. Testing X-SOM: Comparison

16. Conclusions and Future Works

We have realized an ontology mapping tool with high precision and recall. However this task is still not feasible in a fully-automatic fashion.

It seems that the reliability of a matching algorithm is not domain-dependent . The neural network approach works!

Mapping errors often triggered by an questionable ontology design.

Enhancing the recall through the implementation of additional matching techniques .

Replacing the neural network with other machine learning techniques.

Improving the semantic consistency check with more local and global checks.

17. Questions? Further details: [email_address] http://context-addict.elet.polimi.it/

18. Performance Measures:

Precision :

Recall :

F-measure :

WGM :

CM RM CRM

Mapping Space

19. How X-SOM works

X Som Graduation Presentation

Documents