A Case Study for Automatic Query Expansion Based on Divergence

To effectively measure the power of discrimination of terms is a crucial issue in information retrieval (IR). Term mismatch and ambiguity have long been troublesome, and the improvement of retrieval performance by increased precision through query expansion is of interest. This report focuses on a discrimination measure for use with query expansion. The discrimination measure is derived from a divergence measure. A theoretical framework is established. An open problem which arises when applying the divergence measure to query expansion is discussed. The extent of association of terms with the context of the query is defined based directly on the discrimination measure, enabling score functions to be constructed, and good terms for enhancing the query to be identified.