A novel method to enhance recommendation systems via leveraging multiple types of implicit feedbacks

Recommendation system is one of the most widely applied information filtering techniques. In recent years, more and more studies on recommendation systems have shifted from explicit behaviors to implicit behaviors. Although a lot of existing implicit recommendation systems have been proven to have excellent performance, these implicit recommendation systems have two major issues. First, most of studies only consider to utilize one implicit behavior (e.g. click / no click) to learn user preference and help improve recommendation performance. Second, almost all studies neglect the important role of co-occurrence among different implicit behaviors. In this paper, to address the aforementioned challenges, we propose a novel implicit recommendation model Bayesian Personalized Ranking by leveraging Multiple types of Implicit Feedbacks (BPR-MIF), which can distinguish user's favorite degree and make full use of user behaviors. We further leverage the significant role of co-occurrence to highlight implicit behavior combinations that better reflect user preference. In addition, an effective objective function which is suitable for multiple types of implicit behaviors recommendation systems is adopted in our model. And we extend the usage of co-occurrence to a specific item. Ultimately, extensive experiments are conducted on three real-world datasets, including Retailrocket, Douban Book and Jobs datasets. And experimental results have demonstrated that our model outperforms several state-of-the-art implicit recommendation systems in terms of recommendation performance on Retailrocket and Douban datasets.