Collaborative Multi-view Learning with Active Discriminative Prior for Recommendation

Qing Zhang, Houfeng Wang

Abstract: Learning from multi-view data is important in many applications. However, traditional multi-view learning algorithms require the availability of the representation from multi-view data in advance, it is hard to apply these methods to recommendation task directly. In fact, the idea of multi-view learning is particularly suitable for alleviating the sparsity challenge faced in various recommender systems by adding additional view to augment traditional view of sparse rating matrix. In this paper, we propose a unified Collaborative Multi-view Learning (CML) framework for recommender systems, which can exploit task adaptive multi-view representation of data automatically. The main idea is to formulate a joint optimization framework, combining the merits of matrix factorization model and transfer learning technique in a multi-view framework. Experiments on real-life public datasets show that our model outperforms the compared state-of-the-art baselines.