Abstract

Abstract: It is recognized that context plays a crucial role in information retrieval (IR). However, few operational IR systems take it into account. In this paper, we consider one of the important aspects of context - the domain of interest of the user. A query, which is usually short, is interpreted within its domain. This is implemented as the creation of a new statistical language model for the query. The domain is exploited in three different ways: to complete the query, to reorder the retrieval results, or to expand the query using lexical relations extracted from the domain. Our experiments on TREC collections demonstrate clearly the utility of domain for IR. The methods we suggested all lead to significant improvements of retrieval effectiveness.