Query expansion based on clustering and personalized information retrieval

Information retrieval systems are used to describe a variety of processes involving the delivery of information to people who need it. Although several mathematical approaches have been studied in order to formalize the main components of an information retrieval system: queries representation, information items representations and the retrieval process, such systems still face many difficulties to extract relevant information for users especially when the processed data are texts. This is due to the complex nature of text databases. Generally, an information retrieval system reformulates queries according to associations among information items before matching them to dataset items. In this sense, semantic relationships or machine learning techniques can be applied to refine the returned results. This paper presents a formal model to organize data, and a new search algorithm to browse it. It incorporates a natural language preprocessing stage, a statistical representation of short documents and queries and a machine learning model to select relevant results. We propose later in this paper two further optimizations that proved quite interesting and returned significantly satisfying results on two datasets in a reasonable computation time. The first optimization concerns queries expansions, while the second one concerns dataset restructuration. Thus, we formally evaluate the impact of each optimization by computing the performance of the information retrieval system with and without it; the highest reached recall and precision were 96.2% and 99.2%, respectively.