Evidence accumulation with competition in information retrieval

A basic component of a retrieval system is the process of comparing a representation of a document and that of an information need. In this paper, we take the view that information retrieval can be modeled as an inference process and propose a scheme that associatively retrieves potentially relevant documents accumulating evidence from various sources. This scheme that can serve as a precision-improving technique proceeds as follows. It selects an initial subset of promising documents, constructs a network of those selected documents based on their similarities so that they can compete with each other, and accumulate pieces of evidence that a document is relevant, based on multiple criteria so that the documents are ranked. The criteria, by which individual documents are evaluated are currently pragmatic (e.g. importance of terms for the user), semantic (e.g. relationship of terms in query and document), and salience (e.g., term importance in documents). This scheme models the phenomenon of competition among documents for their relevance to a given query. Our experimental results show that this scheme is promising in improving retrieval effectiveness