A Bayesian Personalized Ranking Algorithm Based on Tag Preference

Bayesian Personalized Recommendation algorithm assumes that users prefer the browsed items to the items they have not browsed. This paper identifies users' preferences on items they have not browsed by social tag information. By defining a metric of matching degrees between users and items based on social tags, this paper classifies users' preferences into three types, such as tag-based strong feedback, tag-based weak feedback and negative feedback. We assume that users' preferences for the three kinds of feedback are gradually decreased. A Bayesian personalized ranking algorithm is then proposed based on the preference classification. Experiments on real social tag datasets show that the proposed Bayesian personalized ranking algorithm based on social tags obtains better recommendation results compared with the classical ranking algorithm, especially for sparse data and cold-start recommendation.