Collaborative Filtering based on Modal Symbolic user profiles: Knowing you in the first meeting

Abstract

Recommender systems seek to furnish personalized suggestions automatically based on user preferences. These systems use information filtering techniques to recommend new items which has been classified according to one of the three approaches: Content Based Filtering, Collaborative Filtering or hybrid filtering methods. This paper presents a new hybrid filtering approach getting the better qualities of the kNN Collaborative Filtering method with the content filtering one based on Modal Symbolic Data. The main idea is comparing modal symbolic descriptions of users profiles in order to compute the neighborhood of some user in the Collaborative Filtering algorithm. This new approach outperforms, concerning the Find Good Items task measured by half-life utility metric, other three systems: content filtering based on Modal Symbolic Data, kNN Collaborative Filtering based on Pearson Correlation and hybrid Content-Boosted Collaborative approach. © Springer-Verlag Berlin Heidelberg 2004.