RBPR: A hybrid model for the new user cold start problem in recommender systems

Abstract

The recommender systems aim to predict potential demands of users by analyzing their preferences and provide personalized recommendation services. User preferences can be inferred from explicit or implicit feedback data. Most existing collaborative filtering (CF) methods rely heavily on explicit feedback data. However, these methods perform poorly when rating data is sparse. In this paper, we deal with the extreme case of sparse data, i.e., the new user cold start problem. In order to overcome this problem, we propose a novel CF ranking model, which combines a rating-oriented approach of Probabilistic Matrix Factorization (PMF) and a pairwise ranking-oriented approach of Bayesian Personalized Ranking (BPR) together. Therefore, our proposed model makes full use of the explicit and implicit feedback data. Experiments on the constructed new user cold start datasets based on four public datasets demonstrate the effectiveness of the proposed model for cold start recommendation. Code for the proposed method is available in https://gitee.com/xia_zhaoqiang/recomender-systems-rbpr.