Sparse Gaussian Processes for Learning Preferences

Preference learning has recently gained significant attention in the machine learning community. This is mainly due to its increasing applications in real-world problems such as recommender systems. In this paper, we investigate a Gaussian process framework for learning preferences that uses Expectation Propagation (EP) as its main inference method. This framework is capable of using the collaborative information from all the users for prediction of preferences unlike traditional approaches that only consider single users. We further extend this framework to a sparse setting and show its empirical efficiency. The contribution of this paper is the use of sparse Gaussian process for multi-user preference learning and the comparison of this approach with full EP and the Laplace approximation.