Philip A' Bernstein, Sergey Melnik, John E' Churchill

1
Incremental Schema Matching
Philip A. Bernstein, Sergey Melnik, John E.
Churchill
Microsoft Corporation
What It Does
The Design Space

• Given an element selected in one schema,
  identifies good candidate matches in the other
  schema
• Makes educated guesses, based on name, type,
  structure, and prior matches
• Returns a rank-ordered list that the user
  analyzes to select the desired mapping

Schema-based

• Linguistic
• Lexical
• Acronyms

• Constraints
• Types
• Keys

• Structure
• Nesting context
• Neighborhood

• Reuse-based
• Thesaurus
• Validated matches

• Content-based
• Values
• Value patterns

• Feedback-based
• Action history
• Context

How it Works
Visualization and Navigation

• Do a fast lexical analysis to eliminate
  implausible candidates.
• Score the remaining candidates using a more
  expensive calculation
• Compute a lexical similarity by tokenizing the
  element name and creating weighted
  pseudo-synonyms consisting of ancestors,
  prefixes, vowel-free tokens, and acronyms.
• Add in a weighted structural similarity based on
  nesting context and similarity of element types.
• Compute neighborhood similarity, measured by the
  number of neighbors of the candidate that are
  linked via the current mapping to neighbors of
  the selected element.
• Add a bias to a child when both child and parent
  match. E.g., if Name and its child FirstName are
  both candidates, then FirstName is preferred.
• Display the candidates with the top total
  scores. If one element has the highest score,
  then display it in red.

Screenshot Best match candidate (in the right
pane) is an element in the context of CoSigner
(e.g., not Borrower)

• Usage Scenario
• Select the element to be matched
• Press SHIFT to invoke the matching algorithm
• The best candidate is highlighted
• Down-arrow navigates candidates
• ENTER confirms the current candidate