Automatic query expansion : a comparative study of different strategies for clustering of terms based on local analysis

Automatic query expansion has long been studied in information retrieval research as a technique that deals with the fundamental issue of word mismatch between query and document. The purpose of this thesis is to compare the retrieval effectiveness of different strategies for automatic query expansion. The strategies are based on local analysis of the corpus and use statistical information from the local document set to extract terms that suppose to adapt themselves to each individual search and therefore appear to be searchonyms to the index terms. The strategies compared are: association clusters, metric cluster and scalar cluster. Baseline queries of 24 topics are expanded using terms from the different clusters and searches are made. The study also explores the retrieval effectiveness of an expanded query when using terms derived from the result of a truncation algorithm. The searches were performed in the InQuery IR-system together with the web-based tool QPA and the Swedish database GP_HDINF. The retrieval effectiveness of baseline and the expanded queries are evaluated using relative recall and average precision. The study shows that all of the strategies manage to increase both recall and precision compared with the initial baseline search. No significant differences between the strategies were found.Uppsatsnivå: