Finding Hidden Correlations and Filtering out Incorrect Matchings with Compatibility Detection across Web Query Interfaces

1
Finding Hidden Correlations and Filtering out
Incorrect Matchings with Compatibility Detection
across Web Query Interfaces

• Lei Lei
• June 11, 2004

2
Introduction

• Deep Web scales rapidly
• Proliferating sources with structured Info.
• Vocabulary Converge to small size
• Dynamic Queries instead of URLs

3
Complex Matching

• Traditional methods focus on 11 matching
• Query shemas form Complex Matchings
• Mn

4
Web Query Interfaces

• Web Query Interfaces

Attribute Group
5
Problems to solve

• Relations are complicated and multi-ary
• How to Judge the Relations of Synonyms?
• How to pick out incorrect matchings?

6
Statement

• Find out the hidden synonyms and build
  correlations to solve mn matching problem
• Filter out false matchings and partially
  incorrect ones with the three step compatibility
  detection.

7
MGSsd and Improved Model

• Original Hidden Model from MGSsd

8
Find Hidden Synonyms

• Assume existence of hidden synonyms
• Correlations between synonyms
• Function HC(bi,bj)
• Apply HC directly

9
Example

• Synonyms on air booking domain
• Set a Threshold

HC (b2,b4)
10
Compatibility Detection

• Not all raw matching are correct
• Clean partially correct or inaccurate ones
• Three Steps
• Transitivity Check
• Examine Confidence
• Subsumption

11
Compatibility Detection(Cont.)

• Raw Matching Results
• 1.Check Transitivity

3. Subsumption
2. Choose Confidence
12
Evaluation

• Using Recall and Precision
• Compare with MSGsd data
• Perform Correlation and Compatibility on matching
  results from other researches

13
Contributions

• mn mapping rather than only 11 mapping
• Present a hidden synonym approach to
  statistically compute the correlation between
  synonym groups
• Develop the Compatibility Detection approach to
  refine the raw mapping data
• Suitable and efficient as the Web scales

14
Future Work

• Figure out the HC Function
• Minimum is feasible
• Distinguish Trivial Difference in Confidence
• Set up a proper threshold
• Space Complexity
• Type Subsumption
• Departing datetime
• Departing string

15
Questions ?