Personalized retrieval models aim at capturing user inter-ests to provide personalized results that are tailored to the respective information needs. User interests are however widely spread, subject to change, and cannot always be cap-tured well, thus rendering the deployment of personalized models challenging. In this doctoral work, we describe our approach where we study ranking models from the aspect of search intent. Our approach is query-centric. That is, it does not rely on separate user search profiles/histories nor it personalizes ranking based on topical similarity of queries. Contrary, it examines the search behaviors/intents induced by queries and groups together queries with similar such be-haviors, forming search behavior clusters. Specifically, we exploit user feedback in terms of click data to cluster the queries. Each cluster is finally represented by a single rank-ing model that captures the contained intents expressed by users. Once new queries are issued, these are mapped to the clustering and the retrieval process diversifies possible intents by combining relevant ranking functions. 1.

innovation.gr In this paper, we present a framework for improving the ranking function training and the re-ranking process for web search personalization. Our method is based on utilizing clickthrough data from several users in order to create mul-tiple ranking functions that correspond to different topic areas. Those ranking functions are combined each time a user poses a new query in order to produce a new ranking, taking into account the similarity of the query with each of the topic areas mentioned before. We compare our method with the traditional approaches of training one ranking func-tion per user, or per group of users and we show preliminary experimental results.