Query processing in TREC-6

摘要

Recent improvements in commercial information retrieval systems have focused on query processing as a means to improve the recall and precision of users' queries. This paper discusses three facets of this problem and then describes an attempt to address them so as to improve the retrieval performance. Specifically, this paper considers how techniques for identifying key concepts, selecting synonyms and related terms, and detecting phrases can be combined to expand queries to increase recall and precision. A series of experimental studies are described as performed in the context of the Sixth Text Retrieval Conference (TREC) sponsored by the National Institute of Standards and Technology (NIST). Moderate benefits were achieved through query processing, though additional work needs to be done to fully incorporate phrases, eliminate noise words, and take advantage of co-occurrence analysis developed in this study. Most notably, related terms derived from a statistical thesaurus helped significantly in some queries, while having the opposite effect in others.