Vague Content and Structure VCAS Retrieval for DocumentCentric XML Collections

1
Vague Content and Structure (VCAS) Retrieval for
Document-Centric XML Collections

• Shaorong Liu, Wesley W. Chu, and Ruzan Shahinian
• UCLA Computer Science Department
• sliu, wwc, ruzan_at_cs.ucla.edu
• June 17, 2005

2
Motivation

• XML standard for representation and exchange
• Increasing amount of XML collections available
• Flexible and effective XML retrieval is desired

XML Collections
3
Content Only (CO) Retrieval

• Query
• Content condition what a result should be about
• Example sensor network

4
Strict Content and Structure (SCAS) Retrieval

• Query
• Content condition what a result should be about
• Structure condition what a result is
• Example
• //article/body/sec/paragraphabout(., sensor
  network)

5
Problem

• XML structure is usually very complex
• An XML collection may contain hundreds of tags
• Vague content and structure (VCAS) retrieval is
  more desired
• Fuzzy matching of a querys structure conditions
• Example //article//paragraphabout(., reliable
  multicast)

article
body
sec
sec

paragraph
A comparison of reliable multicast schemes
Uncertainty in sensor network
6
Challenges

• Many existing XML IR engines support
• Content only retrieval and/or
• Strict content and structure retrieval
• How to approximately process structure
  conditions?
• How to measure such fuzzy matching?

7
Related Work

• XML IR engines
• HyREX Fuhr et al., 01
• XXL Theobald et al., 02
• JuruXML Mass et al., 03
• XSEarch Cohen et al., 03
• XRank Guo et al., 03
• Existing approaches to XML VCAS retrieval
• Content-oriented approach e.g., Sigurbjornsson
  et al., 04
• Simple, but lose precision benefits from XML
  structure
• Query-relaxation-based approach Amer-Yahia et
  al., 02 04
• Systematic and efficient, but may miss relevant
  answers

8
Our Approach
VCAS results
VCAS query
SCAS sub-query
SCAS results
Retrieval
Decomposition
Combination
CO sub-query
Retrieval
CO results
9
Roadmap

• Introduction
• Our Approach
• Decomposition
• Retrieval
• Combination
• Experimental Study
• Conclusion

10
Query Language Narrowed Extended XPath I

• Content-oriented XPath-like language Trotman et
  al., 04
• Syntax path1abouts1////pathnaboutsn
• abouts a Boolean combination of about functions
• about function about(path, string)
• Example
• Q1 //articleabout(., sensor network)//paragraph
  about(., reliable multicast)
• Structural hints to a search engine
• Support where to search, e.g., article
• Target what to return, e.g., paragraph

11
Decomposition

• Method
• Example
• Q1 //articleabout(., sensor network)//paragraph
  about(., reliable multicast)
• Q1SCAS //articleabout(., sensor network)
• Q1CO reliable multicast

Q path1abouts1////pathnaboutsn
12
Retrieval

• Use existing XML IR engines to process both
  sub-queries
• Two result lists
• Rscas results for the SCAS sub-query
• Rco results for the CO sub-query

13
Combination

• Method
• Example
• Q1 //articleabout(., sensor network)
  //paragraphabout(., reliable multicast)

v an XML node that is an approximate answer vt
an XML node that matches Qs structure condition
exactly
14
Similarity Measure

• Measure I path-oriented
• Similarity between two nodes is the similarity
  between their corresponding paths from the root
  node
• The greater of common prefixing nodes two paths
  share, the more similar they are

15
Similarity Measure

• Measure II content-oriented
• Similarity between two nodes is the similarity
  between their content
• Use vector space model for content similarity

16
Roadmap

• Introduction
• Our Approach
• Decomposition
• Retrieval
• Combination
• Experimental Study
• Conclusion

17
Experiment Setup

• INEX
• Initiative for the Evaluation of XML retrieval
• Similar to TREC for text retrieval
• Document collection
• Scientific articles from IEEE Computer Society 95
  - 02
• Each article consists of 1500 XML nodes on
  average
• About 500M in size and 170 different XML tags
• Query set all the 33 queries in INEX 04s VCAS
  task
• Gold standard INEX 04 VCAS relevance assessments

18
Experiment Results

• Experiment
• CO (Baseline) ignoring structure conditions in a
  query
• VCAS-1 w/o using similarity measure, i.e., ?vi,
  vj, sim(vi, vj) 1
• VCAS-path using path-oriented similarity measure
• VCAS-cont using content-oriented similarity
  measure
• Results

All the experiments were conducted using our XML
IR engine (Liu et al., 04)
19
Roadmap

• Introduction
• Our Approach
• Decomposition
• Retrieval
• Combination
• Experimental Study
• Conclusion

20
Conclusion

• An approach to XML VCAS retrieval using existing
  engines
• Decomposition
• Retrieval
• Combination
• Experimental results demonstrate the
  effectiveness of the approach
• Future work
• Investigate more general fuzzy matching of a
  querys structure conditions
• Incorporate the efficiency aspect

21
Thank You!
Questions?