Recommender systems provide recommendations to their users for items and services by creating a model tailored to each user to infer their preferences based on previous interactions they have made with the system, as well as possibly using additional information sources. The initial interaction of a user with a recommender system is problematic because, in such a so-called cold start situation, the recommender system has very little information about the user, if any. Moreover, in collaborative filtering, users need to share their preferences with the service provider by rating items while in content-based filtering there is no need for such information sharing. Graph-based user modeling has now become common practice and we have recently shown that a content-based model that uses hypercube graphs can determine user preferences with a very limited number of ratings while better preserving user privacy. In this paper, we confirm these findings on the basis of larger scale experiments.
We thus address the cold start problem by building user models based on a small number of ratings. We also show that the proposed method outperforms standard machine learning algorithms when the number of available ratings is very limited.