This paper introduces a new approach for decision support on the internet. It is characterized by a preference-based filtering relying on the integration of content-based and collaborative filtering principles. We present algorithms that support a key aspect of recommender systems that was absent in early systems: the ability to explain and justify recommendations. A tight integration of preference modeling and machine learning is proposed to support both collaborative decision making and active rating. This integration addresses the problem of both efficiency and storage to scale up with large community of users. For the sake of illustration, we present the main features of the “film-conseil” system which implements the proposed approachfor movie recommendation tasks on the internet.