Extending Faceted Search to the General Web Weize Kong, James Allan CIKM‘14 2014/11/25 (Tue.) Chang Wei-Yuan @ MakeLab Group Meeting

1. +
Extending Faceted Search to the
General Web
2014/11/25 (Tue.)
Weize Kong, James Allan
CIKM‘14
Chang Wei-Yuan @ MakeLab Group Meeting

2. +Outline
n Introduction
n Method
n Facet Generation
n Facet Feedback
n Evaluation
n Experiment
n Conclusion
n Thought
2

3. +Outline
n Introduction
n Method
n Facet Generation
n Facet Feedback
n Evaluation
n Experiment
n Conclusion
n Thought
3

4. +Introduction
n Faceted search helps users by offering
drill-down options as a complement to
the keyword input box.
4

5. +Introduction
n However, this idea is not well explored
for general web search.
n heterogeneous nature
5

6. +Introduction
n However, this idea is not well explored
for general web search.
n heterogeneous nature
6
baggage allowance
所有航線
所有航線
國際航線
國內航線
貨運公司
行李類型

7. +Introduction
n However, this idea is not well explored
for general web search.
n heterogeneous nature
7
baggage allowance
所有航線
所有航線
國際航線
國內航線
貨運公司
行李類型
← query
← facet
← facet term
↓ search result ( ducument)

8. +Introduction
n Goal :
n query-dependent automatic facet generation
n user feedback on these query facets into
document ranking
8

9. +Outline
n Introduction
n Method
n Facet Generation
n Facet Feedback
n Evaluation
n Experiment
n Conclusion
n Thought
9

10. +Flow Chart
10
Search
Result
Candidate
Facets
Facets
Selected
Terms
Top-ranked
Documents
Search
Result
Query
Extracting
Candidates
Refining
Candidates
Facet
Feedback

11. +Flow Chart
11
Search
Result
Candidate
Facets
Facets
Selected
Terms
Top-ranked
Documents
Search
Result
Query
Extracting
Candidates
Refining
Candidates
Facet
Feedback

12. +
Facet
Generation
Facet
Feedback
Evaluation
12
n Input : Query and Search Result
n Step 1 : Extracting Candidates
n Step 2 : Refining Candidates
n Output : Query Facet

13. +
Facet
Generation
Facet
Feedback
Evaluation
13
n Step 1 : Extracting Candidates
n applied both textual and HTML patterns on
the top search results

14. +
Facet
Generation
Facet
Feedback
Evaluation
14
n Step 1 : Extracting Candidates
n query : “mars landing”
n search results
n “ Mars rovers such as Curiosity, Opportunity
and Spirit ”
n candidate facets
n C : { Curiosity, Opportunity, Spirit }

15. +
Facet
Generation
Facet
Feedback
Evaluation
15
n Step 1 : Extracting Candidates
n the candidate query facets extracted.
n noisy
n non-relevant to the issued query
n terms be not members of the same class

16. +
Facet
Generation
Facet
Feedback
Evaluation
16
n Step 1 : Extracting Candidates
n query : “mars landing”
n candidate facets :

17. +
Facet
Generation
Facet
Feedback
Evaluation
17
n Step 1 : Extracting Candidates
n query : “mars landing”
n candidate facets :

18. +
Facet
Generation
Facet
Feedback
Evaluation
18
n Step 1 : Extracting Candidates
n query : “mars landing”
n candidate facets :
n Refine

19. +
Facet
Generation
Facet
Feedback
Evaluation
19
n Step 2 : Refining Candidates
n re-cluster the query facets or their facet
terms into higher quality query facets

20. +
Facet
Generation
Facet
Feedback
Evaluation
20
n Step 2 : Refining Candidates
n Topic modeling
n pLSA, LDA
n Unsupervised clustering method
n QDMiner, QDM
n Super-vised methods based on a
graphical model
n QF-I, QF-J

21. +
Facet
Generation
Facet
Feedback
Evaluation
21
n Input : Query and Search Result
n Step 1 : Extracting Candidates
n Step 2 : Refining Candidates
n Output : Facet : { a set of terms }
n Year : { 2007, 2011, 2012 }
n Lab : { NASA, Mars Science Lab, Curiosity Lab }

22. +Flow Chart
22
Search
Result
Candidate
Facets
Facets
Selected
Terms
Top-ranked
Documents
Search
Result
Query
Extracting
Candidates
Refining
Candidates
Facet
Feedback

23. +
Facet
Generation
Facet
Feedback
Evaluation
23
n Input : Document, Query, User Selection
n Document = one of search result
n Boolean Filtering Model
n Soft Ranking Model
n Output : the score of each document

24. +
Facet
Generation
Facet
Feedback
Evaluation
24
n Boolean Filtering Model
n Fu denotes the set of feedback facets which
user selected
n condition B can be either AND, OR, or A+O
n S(D, Q) is the score returned by the original
retrieval model

25. +
Facet
Generation
Facet
Feedback
Evaluation
25
n Soft Ranking Model
n λ is a parameter for adjusting the weight
n SE(D, Fu) is the expansion part which captures
the relevance between the document and
feedback facet

26. +
Facet
Generation
Facet
Feedback
Evaluation
26
n Input : Documents, Query, User Selection
n Boolean Filtering Model
n Soft Ranking Model
n Output : the score of each document

27. +
Facet
Generation
Facet
Feedback
Evaluation
27
n Intrinsic Evaluation
n Ground Truth: query facets are constructed
by human annotators
n annotators are asked to group or re-group
terms in the pool into preferred query facets.
n pooling facets generated by the different systems
n compared with facets generated by different
systems

28. +
Facet
Generation
Facet
Feedback
Evaluation
28
n Extrinsic Evaluation
n User Model
n The user model describes how a user selects
feedback terms from facets, based on which we can
estimate the time cost for the user.
↑
time for selecting terms
time for scanning facet
↓

29. +
Facet
Generation
Facet
Feedback
Evaluation
29
n Extrinsic Evaluation
n Oracle Feedback and Annotator Feedback
n Oracle feedback model only selected effective terms
as feedback.
n The annotator is asked to select all the terms from
the facets that would help address the information
need.

30. +Outline
n Introduction
n Method
n Facet Generation
n Facet Feedback
n Evaluation
n Experiment
n Conclusion
n Thought
30

31. +Experiment Settings
n Dataset
n For the document corpus, we use the ClueWeb09
Category-B collection.
n 196 queries and 678 query subtopics
n Facet Generation Models
n pLSA, LDA, QDM, QF-I and QF-J
n Facet Feedback Models
n Boolean filtering models, soft ranking models
n Baseline Retrieval Model
n SDM, and its MAP(Mean average precision) = 0.185
31

32. +Facet Generation Models
32

33. +Facet Generation Models
33
based on annotator feedback
and SF feedback model
based on oracle feedback
and SF feedback model.

34. +Facet Generation Models
34
Our experiments testify to the
potential of Faceted Web Search.
based on annotator feedback
and SF feedback model
based on oracle feedback
and SF feedback model.

35. +Facet Feedback Models
35

36. +Facet Feedback Models
36
Our experiments show feedback
models effective.

37. +Outline
n Introduction
n Method
n Facet Generation
n Facet Feedback
n Evaluation
n Experiment
n Conclusion
n Thought
37

38. +Conclusion
n This paper proposed Faceted Web
Search.
n an extension of faceted search to the general
Web
n query-dependent automatic facet
generation
n feedback on these query facets into
document ranking
38

39. +Outline
n Introduction
n Method
n Facet Generation
n Facet Feedback
n Evaluation
n Experiment
n Conclusion
n Thought
39

40. +
Thanks for listening.
2014 / 11 / 25 (Tue.) @ MakeLab Group Meeting
v123582@gmail.com