Vers l’utilisation de relations de préférence pour le filtrage collaboratif

Collaborative filtering based recommender systems exploit users preferences about items to provide recommendations to these users. These preferences are generally ratings. However, choosing a rating is not an easy task for any user ; the rating value may be influenced by many factors and the ratings are thus not completely trustworthy. In this article, we propose a new approach of expressing preferences, under the form of preference relations instead of ratings. Similar steps as in classical collaborative filtering are used and we propose new measures to exploit preference relations to compute recommendations. First experimentaions of this new approach have been conducted on a state of the art corpus of the recommender systems domain.