Query Mapping Across Heterogeneous Information Sources

1
Query Mapping Across Heterogeneous Information
Sources

• Paper Review

2
Introduction

• Mediator
• Translate queries from unified context to native
  context
• Query a boolean expression of constraints
• attr op value or attr1 op attr2
• Mapping map attributes convert data values
  transform operators
• Approach
• Rule-based framework for necessary domain
  semantics need human expertise
• An algorithm for optimal mapping new query
  minimally subsume old query
• Address constraints dependency, arbitrary
  constraints, semantic mapping, complex
  queries(with conjunctions and disjunctions)

3
Simple-conjunction queries

• Rule-based
• translate the smallest grouping of basic
  constraints
• Independent and not recursive
• Sample lnLfnF gtALnFnToName(L,F)
  emitauthor A
• Algorithm SCM (Simple-Conjunction Mapping)
• Input Q a simple query K mapping rules for
  target T
• Output S(Q) minimal subsuming mapping w.r.t. T
• Procedure find matchings by rules remove
  redundant matching output S(Q)

4
DNF-based schema for Complex Queries

• Algorithm DNF
• Input Q an arbitrary query K mapping rules
  for target system T
• Output S(Q) minimal subsuming mapping w.r.t. T
• Procedure convert Q into DNF apply SCM return
  S(Q)
• Problems
• Expensive, inflexible, sometime unnecessary
• Exponential in the number of constraints

5
Traversal-based Top_Down Query Mapping

• Representation
• Query tree node, leaf constraints
• TDQM algorithm
• Input Q arbitrary query K mapping rules for T
• Output S(Q) minimal subsuming mapping
• Procedure partition Q into blocks, then SCM

6
Conjunct Seperation

• Psafe algorithm

7
Optimality, compactness, and complexity

• Optimality
• Basic rules codify human expertise
• Correctly handle conjunctions
• Compactness
• TDQM generates more compact mappings vs.
  DNF-based
• Rewrite query only when necessary

8
Conclusion

• Algorithms produce both optimal and compact
  mappings
• Efficient time linear to input size

9
Related to our work

• Pro
• Human defines rules similar to our defining XML
  specification based on DTD
• Algorithms are somehow helpful we need to define
  our own algorithm for mapping
• We can work on handling more complex mapping
• Con
• Not quite fit for DL, which is more organized
• Our approach has little control over target
  system, just mapping based on outside behavior,
  little knowledge on constraints

10
(No Transcript)