Leveraging Multi-Actions to Improve Bayesian Personalized Ranking for Heterogeneous Implicit Feedback

Implicit feedback (e.g., purchases, views, clicks, follows, listens, likes) have been receiving more attention due to their close relationship with real applications. In some recent works, pairwise ranking algorithms have been shown to leverage multiple types of auxiliary implicit feedback for solving the recommendation problem of heterogeneous implicit feedback (HIF). In this paper, we develop a novel preference learning algorithm to learn a confidence over multiple users’ actions, which is called Multi-actions Auxiliary Feedback Confidence Learned Bayesian Personalized Ranking (MAC-BPR). MAC-BPR has the merits of generating effective data from multiple auxiliary feedback and accommodating it for accurate pairwise preference learning. Experimental results on the real-world dataset demonstrate that MAC-BPR achieves better recommendations than several state-of-the-art methods over various evaluation metric.