Collaborative filtering matches the user based on their similar tastes or preferences and suggests items of their choice. This paper presents a new probabilistic model for collaborative filtering, namely real preference Gaussian mixture model (RPGMM) which has two novel features when compared to the existing models and algorithms. First, it has two latent classes for users and items. Each user or item may be probabilistically clustered to multiple groups. Second, the model considers the user's rating style and item admiration. Therefore the rating can be structured to a Gaussian mixture model with three factors. They are the user's real preference to the item, the rating style of the user, and the public praise of the item. Experiments on EachMovie dataset show that the new model outperforms several other collaborative filtering models and algorithms remarkably.
Author's Name: Zhang, L., Li, M.-Q.
Volume: Volume 1
Issues: Issue 4
Keywords: Collaborative filtering, Gaussian mixture model, Latent variable model