HLBPR: A Hybrid Local Bayesian Personal Ranking Method

Bayesian Personal Ranking(BPR) method is a well-known model due to its high performance in the task of item recommendation. However, this method fail to distinguish user preference among the non-interacted items. In this paper, to enhance traditional BPR's performance, we introduce and analyse a hybrid method, namely Hybrid Local Bayesian Personal Ranking method(HLBPR for short). Our main idea is to construct additional item preference pairs among the products which haven't been purchased, and then utilize the extened pairs to optimize the ranking object. Experiments on two real-world transaction datasets demonstrated the effectiveness of our approach as compared with the state-of-the-art methods.