Yi Zhang, Jiazhong Nie
07/16/2010 09:00 AM
Applied Mathematics & Statistics
Many recommender systems might be part of an e-commerce or multi functional system (or portal) where various information about users, products/documents, social networks, and different types of user feedback about products/documents are available. This paper exploits the heterogeneous information a recommender system might collect to make the most appropriate recommendations. We propose a new Probabilistic Latent Relational Modeling (PLRM) approach to jointly model user, product, social network, and different types of implicit and explicit user feedback. A system can make different types of recommendations, such as who you may want to trust or what product you may want to buy, based on a single probabilistic model. Different types of user feedback collected can be utilized to improve the performance of recommendations. We also propose a hybrid variational Bayesian and Max A Posteriori method to estimate the parameters from various types of user feedback. The experimental results on two Epinion.com data sets demonstrate the effectiveness of the proposed modeling approach.
