Using a concept-based user context for search personalization

Because of the diversity of the user interests and the ambiguity of the user query, current search engines are not very effective. Indeed, they are based on simple query-document matches without considering the user background and interests. Personalized search aims at integrating the user context, defined as a set of user's topics of interests, in the information retrieval (IR) process in order to tailor search results to a particular user. An effective personalization is achieved when an accurate representation of the user context is provided. We present in this paper our approach for learning long term user interests by collecting information from the user's feedback and using existing domain ontology. The learning process is based on the aggregation of the short term user contexts represented as a set of general concepts, where the user context reflect the user's topics of interest in a specific search session. Personalization is achieved by using the user contexts across related search sessions. Our experimental results carried out in TREC collection show that re-ranking the search results based on the concepts weights of the short term user context brings significant improvements in the retrieval precision.