A straightforward generalization of low rank approximation approach for hybrid recommender systems

We propose a new hybrid approach for matrix- and tensor-based recommender systems. Unlike the majority of hybrid recommenders, it directly ties collaborative user behavior with additional side information in an intuitive and straightforward way. It not only helps to address the problem of extreme data sparsity, but also allows to naturally exploit patterns in the observed interactions for constructing a compact and meaningful representation of user intents. We demonstrate the effectiveness of the proposed model on several standard benchmark datasets. The general formulation of the approach imposes no restrictions on the type of observed interactions and makes it potentially applicable for joint modelling of any type of contextual information along with side data.