Federated Multi-view Matrix Factorization for Personalized Recommendations

We introduce the federated multi-view matrix factorization method that extends learning framework to with multiple data sources. Our is able learn model without transferring user's personal a central server. As far as we are aware this first provide recommendations using factorization. The rigorously evaluated on three datasets production settings. Empirical validation confirms outperforms simpler methods do not take into account structure of data, in addition, it demonstrates usefulness proposed for challenging prediction tasks cold-start recommendations.