Computer Science
@WPI
Relevance Ranking for Web Metasearch
Student:
  Anna Novikov
Advisors:
  Carolina Ruiz
  Sergio A. Alvarez
|
Scope
This project focuses on mechanisms that allow the amalgamation
of the ranked hit lists provided by typical search engines in
response to user queries, into summary lists that can be used to produce
the final list output by a metasearch engine. The techniques used are
based mainly on the theory of preference voting. Search engines are viewed
as voters in an election in which the winning candidate is the
top-ranked hit for a given user query; more generally, the support
given by the voting engines to the various candidates (hits)
produces the resulting ranking.
Three voting-theoretic methods are compared: Condorcet, Borda count,
and plurality with elimination. In addition, a technique based on
geometric transformations is compared with the above methods.
A prototype system incorporating the above methods has been built.
The system adapts to user preferences by maintaining a metaindex
that is updated dynamically based on the frequency with which
the various search engines have provided links that the user has
subsequently followed.
|