Abstract:

The implicit feedback modeling user-Point-of-Interest（POI）interaction accuracy of the POI recommendation system is not high and the implicit feedback attribute of the use’s check-in data is ignored. A novel POI recommendation algorithm is proposed. Specifically, first of all, a neural network-based ranking algorithm is used to capture the interaction relationship of user-POI. Then, the Poisson factorization algorithm and Bayesian personalized ranking technology are combined to model the user’s check-in behavior. The algorithms obtained in the above two steps are integrated into a unified recommendation algorithm architecture to provide a POI recommendation service. The experimental results show that the proposed algorithm is better than the traditional state-of-the-art POI recommendation algorithm.

Key words: Point-of-Interest（POI） recommendation, Poisson factorization, neural network, Bayesian personalized ranking

摘要：

针对兴趣点推荐系统存在的隐式反馈建模用户-POI交互准确率不高和忽视用户签到数据的隐性反馈属性的问题。提出了一种新颖的兴趣点推荐算法。具体而言，采用一种基于神经网络的排序算法来捕获用户-兴趣点的交互关系，结合泊松分解算法和贝叶斯个性化排序技术建模用户的签到行为，将上述2个步骤得到的算法整合到统一的推荐算法架构中，从而提供兴趣点推荐服务。实验结果表明，提出的算法推荐性能优于传统主流先进兴趣点推荐算法。

关键词: 兴趣点推荐, 泊松分解, 神经网络, 贝叶斯个性化排序