Your SlideShare is downloading. ×
  • Like
Dexa2007 Orsi V1.5
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×

Now you can save presentations on your phone or tablet

Available for both IPhone and Android

Text the download link to your phone

Standard text messaging rates apply

Dexa2007 Orsi V1.5

  • 250 views
Published

 

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads

Views

Total Views
250
On SlideShare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
3
Comments
0
Likes
0

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide
  • Ricerca di sorgenti
  • Ricerca di sorgenti
  • Ricerca di sorgenti

Transcript

  • 1. X-SOM A Flexible Ontology Mapper Carlo Curino, Giorgio Orsi , Letizia Tanca {curino,orsi,tanca}@elet.polimi.it Politecnico di Milano Dipartimento di Elettronica e Informazione September 4 th SWAE 2007 (DEXA’07)‏ Regensburg
  • 2. Motivations
    • Part of the Context-ADDICT Project (Context Aware Data Design Integration Customization and Tailoring).
    • Scenarios:
      • Ontology-based integration of heterogeneous data sources
      • Semantic Web applications
      • Knowledge Management
    • Tasks:
      • Semantic (semi-)automatic ontology matching/mapping/aligning…
      • Semantic consistency checking
  • 3. Outline
    • Problem setting.
    • X-SOM algorithm.
    • From matchings to mappings: The debugging process.
    • Experimental Results.
  • 4. The Problem Alignment Ontology Alignment : The process of bringing two or more ontologies into mutual agreement , by relating their constitutive elements by means of alignment relationships , and making them coherent and consistent. .
  • 5. The Problem Matching Ontology Alignment : The process of bringing two or more ontologies into mutual agreement , by relating their constitutive elements by means of alignment relationships , and making them coherent and consistent. .
  • 6. The Problem Mapping Ontology Alignment : The process of bringing two or more ontologies into mutual agreement , by relating their constitutive elements by means of alignment relationships , and making them coherent and consistent.
  • 7. X-SOM’s mapping process Matching: Similarities between ontologies computed with a customizable set of matching algorithms (strategy). The results are combined by means of a feed-forward neural network. Debugging: Matchings are tested for consistency and coherency to improve their quality. Conflicts are solved in a (semi-)automatic fashion. Mapping: An ontology containing the mappings between the constitutive components of the input ontologies.
  • 8. X-SOM Architecture
    • Three Subsystems:
      • Matching
      • Mapping
      • Inconsistency Resolution
  • 9. Matching phase: Production
    • Features:
      • Syntactic (Jaro, Levenshtein, …), structural and semantic (WordNet, Google, …) similarities.
      • A module can use other modules results to have a starting point for its algorithm (e.g., structural ones).
      • X-SOM matching modules are designed to exploit intrinsic parallelism of matching algorithms where possible.
    • Where are the problems?
      • The optimal combination function is often non-linear: It is approximated via machine learning .
      • Matching strategy definition: What modules are suitable for a given mapping task?
  • 10. Matching phase: Combination
    • X-SOM’s Neural Network:
      • X-SOM combines the modules’ outputs using a three-layers feed-forward neural network.
      • Training set built from data (benchmarks ontologies).
      • The Neural network increases performance up to 15% in precision and 35% in recall if compared with simple average functions (LWM, QWM, sigmoid, etc.).
    • Controversial points:
      • Is the learned function domain dependent?
      • How to build a good training set?
  • 11. Controversial points
    • Domain Independence:
      • Learned function robust to domain changes, but
      • It is not robust to different design techniques.
      • The network learns the intrinsic reliability of the matching algorithms (and their combinations).
    • Training set:
      • The number of samples with positive and negative outcomes must be balanced.
      • The techniques influence each others: selection of almost independent techniques.
  • 12. Matchings debugging
    • Semantic consistency checking :
    • The process of verifying whether there are mappings that modify the semantics of the elements belonging to the original ontologies.
    • Debugging process:
      • Guarantees satisfiability while preserving the semantics of the original ontologies.
      • Makes use of heuristics and of an extended tableau algorithm for description logics to allow matching debugging and explanation.
      • Addresses multiple mappings .
  • 13. Semantic consistency: Examples
    • Bowties:
    • Cycles:
  • 14. Semantic consistency: Solutions
    • Bowties:
    • Cycles:
  • 15. Experimental Results: OAEI 2007
  • 16. Experimental Results: OAEI 2007
  • 17. Conclusion and Future Work
    • Summary :
      • We presented an extensible ontology mapper that combines several matching algorithms by means of a neural network and uses a debugging process to improve the quality of ontology mappings as well as guarantee the consistency of the mapping.
      • We tested its performance against the OAEI’07 benchmarks.
    • Future Work:
      • Increase mappings expressiveness (Heterogeneity / GLAV).
      • New modules: e.g., pure structural matchers, instance and instance-based matchers.
      • How can collaborative background knowledge improve mapping algorithms?
  • 18. Question time
    • Q & A
    • (If I’m showing this slide, I haven’t run out of time)‏
  • 19.  
  • 20. Overall System Architecture
  • 21. Models view
  • 22. Data Tailoring
    • Data Tailoring, based on the Dimension Tree Instantiation :
    • Schema Tailoring
    • Instance Tailoring
  • 23. Semantic Extraction
    • Data Source Ontology:
    • Semantic Extraction: data abstract model + storage model
    • Supports the query processing
    • Models isolation (different models can be used separately)‏