Query Expansion and Classification of Retrieved Documents

This paper presents different methods tested by the University of Avignon and Bertin at the TREC-7 evaluation. A first section describes several methodologies used for query expansion: synonymy and stemming. Relevance feedback is applied both to the TIPSTER corpora and Internet documents. In a second section, we describe a classification algorithm based on hierarchical and clustering methods. This algorithm improves results given by any Information Retrieval system (that retrieves a list of documents from a query) and helps the users by automatically providing a structured document map from the set of retrieved documents. Lastly, we present the first results obtained with TREC-6 and TREC7 corpora and queries by using this algorithm. keywords: ad-hoc information retrieval, automatic relevance feedback, synonymy, automatic classification, cluster-based and hierarchical methods.