Answering Imprecise Queries over Autonomous Web Databases By Ullas Nambiar and Subbarao Kambhampati

1
Answering Imprecise Queries over Autonomous Web
DatabasesBy Ullas Nambiar and Subbarao
Kambhampati

• Anthony Okorodudu
• CSE 6392
• 2006-4-11

2
Outline

• Introduction
• Overview
• AIMQ System
• Approach
• Attribute Ordering
• Query-Tuple Similarity
• Conclusion

3
Introduction

• Database query processing models assume user
  knows what they want and how to formulate query
• Users can tell which tuples are of interest to
  them
• Domain and user independent solution for
  supporting imprecise queries over autonomous Web
  databases

4
Overview

• Example Suppose a user wishes to search for
  sedans priced around 10,000 in a used car
  database.
• Table Schema CarDB(Make, Model, Year, Price,
  Location)
• Query CarDB(Model Camry, Price lt 10000)

5
Overview (continued)

• Since Accords are similar, user may also be
  interested in these
• User may also be interested in price slight above
  10,000
• Basic query processing will not return tuples not
  specifically satisfying query
• User will have to manually issue queries for all
  similar models

6
Overview (continued)

• Automate by telling query processor information
  about similar models
• Difficult to specify domain specific similarity
  metrics

7
AIMQ

• Remove burden of providing value similarity
  functions and attribute orders from users
• Attempt to reduce human input needed for
  satisfactory answer

8
AIMQ Approach

• Query CarDB(Model like Camry, Price like 10000)
• Base Query
• Qpr CarDB(Model Camry, Price 10000)
• Assume non-null resultset
• Sample result
• MakeToyota, ModelCamry, Price10000, Year2000
• Issue queries relaxing any of the attribute
  bindings

9
AIMQ Approach (continued)

• Which relaxations will produce more similar
  tuples?
• How to compute similarity between the query and
  an answer tuple?
• Ans(Q) x x ? R, Similarity(Q,x) gt Tsim

10
Attribute Ordering

• Tuples most similar to t will differ only in the
  least important attribute
• Identifying least important attribute
  necessitates an ordering of attributes in terms
  of their dependence on each other
• Estimate importance of attribute by learning the
  Approximate Functional Dependency (AFD) from a
  sample of the database

11
Attribute Ordering

• Use Approximate Functional Dependency (AFD) to
  create attribute dependence graph
• Remove cycles and partition into dependent and
  deciding set
• Relax members of dependent sets ahead of deciding
  set

12
Attribute Relaxation Order
13
Categorical Value Similarity

• Similarity between two values binding a
  categorical attribute, VSim, is the percentage of
  common Attribute-Value pairs that are associated
  to them
• Tuple Ford, Focus, 15k, 2002
• AV-pair MakeFord is associated to the AV-pairs
  ModelFocus, Price15k, and Year2002

14
Categorical Value Similarity
15
Categorical Value Similarity

• Measure similarity between two AV-pairs as the
  similarity shown by their supertuples

16
Categorical Value Similarity
17
Conclusion

• AIMQ is a domain independent approach for
  answering approximate queries over autonomous
  databases
• Attempt to reduce human input needed for
  satisfactory answers

18
References

• U. Nambiar and S. Kambhampati. Answering
  Imprecise Queries over Autonomous Web Databases.
  ICDE Conference.

19
Thanks