1. Weiyi Ge, Gong Cheng, Huiying Li, Yuzhong Qu
Incorporating Compactness to Generate Term-Association View Snippets for Ontology Search
Weiyi Ge, Gong Cheng, Huiying Li, Yuzhong Qu
Accepted by Information Processing and Management

2. Snippet for Ontology Search
Why need ontology search?
Reuse
Amount
Why need snippet?
Numerous ontologies
Numerous triples/terms
Reuse: enhance data interoperability, reduce cost and accelerate development progress

3. Requirements for Snippets
1. Semantic relations among terms are preferred

4. Requirements for Snippets
2. Direct relations or indirect but close relations are preferred

5. Requirements for Snippets
3. Compact

6. Requirements for Snippets
Term association view
Semantic relations among terms are preferred
Direct relations or indirect but close relations are preferred
Compact
Bounded size
Query relevant
In summary,
A good snippet for ontology search is a term-association view information unit within a bounded size that effectively summarizes the query result for human reading and judgment.

7. Outline Semantic relations among terms  term association graph (TAG)
Close relations  maximal r-radius subgraph
Compactness  group Steiner problem
Size+relevance  assembly
Evaluation

8. Outline Semantic relations among terms  term association graph (TAG)
Close relations  maximal r-radius subgraph
Compactness  group Steiner problem
Size+relevance  assembly
Evaluation

9. Term Association Graph
RDF Sentence
it makes no sense if RDF triples sharing common blank nodes are separated

10. Term Association Graph
X(t1, t2, R) is a set of RDF sentences in R, in each of which there is a directed path connecting t1 and t2 whose arcs exclude rdf:type.
Preference
Preference of sentence: prefS(S)
Preference of term association: prefX(X(t1,t2,R))=ΣprefS(S)

11. Term Association Graph

12. Term Association Graph

13. Outline Semantic relations among terms  term association graph (TAG)
Close relations  maximal r-radius subgraph
Compactness  group Steiner problem
Size+relevance  assembly
Evaluation

14. Maximal r-Radius Subgraph
Motivation
Relations with close distances
Size restriction
Reduce the graph scale
Defintion
r-Radius Subgraph
Gi is a subgraph of G and the radius of Gi is less than or equal to r.
Maximal r-Radius Subgraph
No other r-Radius subgraph is its supergraph.

15. Maximal r-Radius Subgraph

16. Maximal r-Radius Subgraph

17. Outline Semantic relations among terms  term association graph (TAG)
Close relations  maximal r-radius subgraph
Compactness  group Steiner problem
Size+relevance  assembly
Evaluation

18. Sub-snippet Generation
Given a maximal r-radius subgraph Gi and a set of relevant keywords Qi = {q1;…;qg}, a sub-snippet Gsub is a connected subgraph of Gi satisfying qQi, vV(Gsub) such that qLblV(v).
Compact sub-snippet

19. Sub-snippet Generation

20. Sub-snippet Generation

21. Outline Semantic relations among terms  term association graph (TAG)
Close relations  maximal r-radius subgraph
Compactness  group Steiner problem
Size+relevance  assembly
Evaluation

22. Snippet Generation Requirements
Compactness
Relevance
Quality = *Compactness + (1-)*Relevance
Snippet S
max(Quality(GS))
size(GS)<=restriction
Greedy method
maximal marginal increase

23. Snippet Generation

24. Evaluation Feasibility 4,522 ontologies
two 4-core Xeon E7400 (2.4G) and 24GB memory

25. Evaluation Effectiveness Competitors Participants Experimental Design
Term set (Term.S, Term.W)
Sentence set (Sent, Sent+Q)
Term Association (TA+C, TA)
Participants
30 volunteers
Experimental Design
10 topics from ODP
Questionnaire

26. Evaluation

27. Any questions welcome