Multicriteria collaborative filtering by Bayesian model-based user profiling

This paper proposes a Bayesian model for multicriteria (MC) recommender systems, which are useful tools for delivering information to those who require it. Such systems usually handle a single overall rating score to capture user´s preferences. Recently proposed MC recommender systems use multiple scores evaluated from various aspects to obtain a more elaborate user profile. Our proposed model maps users and items to their groups via corresponding latent topics. We empirically evaluated the proposed model and showed that: (1) the Bayesian model is effective in estimating many parameters required for MC recommendation models; and (2) the multinomial distribution, which is usually used in latent models, is insufficient for predicting absolute rating scores.