Enhancing Collaborative Filtering Recommendation by User Interest Probability
Jing Yu, Jingjing Shi, Yunwen Chen, Wenhai Liu, Kai Liu, Zhijun Xie · 2021
Traditional collaborative filtering recommendation is one of the most commonly used algorithms in current recommendation systems, and it is also the mainstream algorithm used in the e-commerce industry. The basic principle of collaborative filtering is that users with similar interests will have similar interest bias in the future. However, the traditional collaborative filtering algorithm faces some of the following problems in the e-commerce industry: (1) lacking of confidence based on user interest bias similarity; (2) without consideration of the time factor; (3) no correlation between behaviors. Therefore, the collaborative filtering model based on user interest probability (PUCF) proposed in this paper to solves the above problems. Firstly, Wilson confidence interval is used to solve the problem of confidence, the behavioral time decay factor is obtained through the normalized time. And then considering the problem of conversion between behaviors. Besides the algorithm assigns different calculation weights to the existence of progressive behavior relationships, and to a certain extent considers the influence of the internal connections between behaviors on user interest. Through looking for the optimal parameters and making comparative experiments, it shows that the effect of collaborative filtering model based on user interest probability (PUCF) proposed in this paper is better than other collaborative filtering methods.