1. Gong Cheng, Yanan Zhang, and Yuzhong Qu
Explass: Exploring Associations between Entities via Top-K Ontological Patterns and Facets
Gong Cheng, Yanan Zhang, and Yuzhong Qu

2. Contents Introduction Association Definition Overview of Explass
Approach
Evaluation
Conclusion
Next work

3. Introduction
What are the associations between A and B ?

4. Introduction How to efficiently find associations.
How to help users explore a large set of associations that have been found.

5. Introduction (Related Work)
Association Discovery and Ranking
Exploratory Association Search

6. Introduction Our work:
provides a flat list (top-K, rather than a hierarchy) of clusters for refocusing.
mine all the significant patterns
find top-K ones that are as frequent and informative as possible while sharing small overlap between each other.
integrates patterns with facet values.

7. Association Definition
Association (Path-based)
G = <V, A, s, t, lV, lA>, a path v0a1 · · · anvn from eS = lV (v0) to eE = lV (vn): Z= r1e1 · · · en−1rn,
for 1 ≤ i ≤ n − 1, ei = lV (vi), and
for 1 ≤ i ≤ n, if s(ai) = vi−1, then ri = lA(ai); otherwise, ri = ˜lA(ai).
Ontological association pattern
P = r’1c’1· · · c’n−1r’n, denoted by Z ∈ M(P)
for 1 ≤ i ≤ n − 1, ei ∈ I(c’i), and
for 1 ≤ i ≤ n, ri ⊑R r’i.

8. Association definition
Alice
Bob
PaperC
PaperB
secondAuthor
inProcOf
cites
ConfB
extends
PaperA
PaperD
ArticleA
ConfA
reviewer
firstAuthor
chair

9. Overview of Explass Filters in use
Facet values
(classes)
(relations)
Click to use this pattern as a filter
Associations matching a recommended pattern
Associations not matching any recommended pattern

10. Pattern Recommendation
Mining Signicant Patterns
To characterize the relevance of pattern P to the query context,
secondAuthor
Author
RELATED
psc(PaperA)
ConfPaper
Publication
ENTITY …
2/5
1/5

11. Pattern Recommendation
Mining Signicant Patterns
Data mining
Frequent closed itemset mining problem(FCIMP)
Encode the path structure
Association->transaction
Item: a position-relation pair {1, 3, … , 2n − 1} × ∑R or a position-class pair in {2, 4, … , 2n − 2} × ∑C

12. Finding Frequent, Informative, and Small-Overlapping Patterns
Informativeness
self-information (specific)
entropy

13. Finding Frequent, Informative, and Small-Overlapping Patterns
ontological overlap
contextual overlap
Optimization
find up to K ones that are as frequent and informative as possible while sharing small overlap between each other
Multidimensional 0-1 knapsack problem (MKP)
Greedy heuristic
P = r1c1· · · cn−1rn
P’ = r’1c’1· · · c’n−1r’n
hits(p)
hits(p’)

14. Facet Value Recommendation
K classes of entities and K relations
frequency
Informativeness
overlap

15. Evaluation
To investigate how patterns and facets help users explore associations in practice
Two hypotheses
H1. For association exploration, providing a flat list (top-K) of frequent, informative, and small-overlapping patterns (as on Explass) is more satisfying than an inclusive hierarchy of patterns (as on RelClus).
H2. Patterns and facets are notably complementary in terms of usage in association exploration, and thus providing both of them (as on Explass) is more satisfying than only one of them (as on RelFinder and RelClus).

16. Evaluation Data Sets: DBpedia Tasks
Derived from the 100 training queries (QALD-3 evaluation campaign)
Related entities that “people search for” by Goolge Search
26 tasks
Explass vs RelClus vs RF (reproduced RelFinder)

17. Results
User Experience

18. Results
User Behavior

19. User Feedback and Discussion
RelClus
6 subjects(30%):
provided a good overview of all the associations and helped refocus on a particular theme
11 subjects(55%):
a high level were often too general to be useful, confused about the deep and complicated hierarchies RF
5 subjects (25%)：
recommended classes and relations were useful filters
8 subjects (40%)：
needed a better overview for summarizing associations

20. User Feedback and Discussion
Explass
14 subjects (70%):
provided a good summary of associations and helped refocus on a particular theme when recommended facet values helped filter associations
11 subjects (55%):
some very large clusters could be divided into small ones
As to H1: Explass considered the informativeness of patterns in recommendation.
As to H2: patterns provided an overview that meaningfully summarized significant subsets of associations covering diverse themes to be refocused on, when facets provided useful filters for refining the search.

21. Conclusion
realized exploratory association search in a new way by recommending Top-K patterns and facet values, which have been shown to be notably complementary in terms of usage: patterns for summarizing and refocusing, and facets for refining and filtering.

22. Next work Discover implicit semantic associations between entities
Other types of associations
Type similarity
<p, l> similarity …
Named associations with understandable and meaningful labels.
Virtual properties
Semantic metrics
Ranking (pruning)

23. References
Aleman-Meza, B., Halaschek-Wiener, C., Arpinar, I.B., Ramakrishnan, C., Sheth, A.P.: Ranking Complex Relationships on the Semantic Web. IEEE Internet Comput. 9(3), 37– 44 (2005)
Anyanwu, K., Maduko, A., Sheth, A.: SemRank: Ranking Complex Relationship Search Results on the Semantic Web. In: 14th International Conference on WorldWide Web, pp. 117–127. ACM, New York (2005)
Anyanwu K, Sheth A. Ρ-Queries: enabling querying for semantic associations on the semantic web[C]//Proceedings of the 12th international conference on World Wide Web. ACM, 2003:
Jeh G, Widom J. SimRank: a measure of structural-context similarity[C]//Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2002:
Araujo S, Houben G J, Schwabe D, et al. Fusion–Visually Exploring and Eliciting Relationships in Linked Data[M]//The Semantic Web–ISWC Springer Berlin Heidelberg, 2010: 1-15.
Lassila O. Generating Rewrite Rules by Browsing RDF Data[C]//Rules and Rule Markup Languages for the Semantic Web, Second International Conference on. IEEE, 2006:

24. Thanks