Affiliations: Université de Toulouse, Institut de recherche en informatique de Toulouse SIG-RI, 118, route de Narbonne, 31062 Toulouse (France) CEDEX 09, France
Abstract: It is well known that a query is an approximate representation of the user's information needs since it does not provide a sufficient specification of the attended results. Numerous studies addressed this issue using techniques for better eliciting either document or query representations. More recent studies investigated the use of search context to better understand the user intent, driven by the query, in order to deliver personalized information results. In this article, we propose a personalized information retrieval model that leverages the information relevance by its usefulness to both the query and the user's profile, expressed by his main topics of interest. The model is based on the influence diagram formalism which is an extension of Bayesian networks dedicated to decision problems. This graphical model offers an intuitive way to represent, in the same framework, all the basic information (terms, documents, user interests) surrounding the user's information need and also, quantify their mutual influence on the relevance estimation. Experimental results demonstrate that our model was successful at eliciting user queries according to dynamic changes of the user interests.
Keywords: Personalized information retrieval, influence diagrams, aggregation operators, decision-making
