Evaluation of density one-class classifiers for item-based filtering

In this paper we explore the use of density one-class classifiers for the movie recommendation problem. Our motivation lies in the fact that users of recommender systems usually provide ratings only for items that they are interested in and belong to their preferences without giving information on items that they dislike. One-class classification seems to be the proper learning paradigm for the recommendation problem, as it tries to induce a general function that can discriminate between two classes of interest, given the constraint that training patterns are available only from one class. The experimental results show that one-class classifiers not only cope with the problem of lack of negative examples, but also succeed in performing efficiently in the recommendation process.