Feature subspace transfer for collaborative filtering

The sparsity problem is a major bottleneck for the collaborative filtering. Recently, transfer learning methods are introduced in collaborative filtering to alleviate the sparsity problem which aim to use the shared knowledge in related domains to help improve the prediction performance. However, most of the transfer learning methods assume that the user features or item features learned from different data matrices have the same dimensions which is often not met in practice. In this paper, we propose a transfer learning method for collaborative filtering, called Feature Subspace Transfer (FST) to overcome this limitation. In our model, the user feature subspace learned from the auxiliary data is transferred to the target domain. An iterative algorithm is also proposed for solving the optimization problem. Numerical experiments on real-world data show the improvement of our method on alleviating the sparsity problem.