Abstract:

POI（Point-Of-Interest） recommendation for location-based social network exist some problems such as sparse user check-in matrix, low recommendation accuracy and inadequate use of context information. This paper proposes a matrix decomposition algorithm TGMF to solve the above problems. First, it uses BPR model to optimize the process of matrix decomposition and improves the generation strategy of partial order relations. Secondly, it fuses the trust effect and the similarity computation to improve the accuracy of recommendation. Finally, two models are combined to get the final preference list of users, and top-[k] interest points are recommended to users. Experimental results show that on real data sets Gowalla and Foursquare, the proposed algorithm is superior to the traditional interest point recommendation algorithm in precision and recall.

Key words: Point-Of-Interest（POI） recommendation, matrix factorization, social trust, BPR

摘要：

基于LBSN的兴趣点推荐存在用户签到矩阵稀疏、推荐精度不高、上下文信息利用不充分等问题，提出一种融合社交信任的矩阵分解算法TGMF（Trust-Geo?Matrix?Factorization）来缓解以上问题。利用BPR模型优化矩阵分解的过程，改进偏序关系的生成策略。把信任影响和相似度计算相结合，提高推荐精度。融合两种模型得到用户的最终偏好列表。把偏好列表中的top-[k]个兴趣点推荐给用户。实验结果表明，在真实数据集Gowalla和Foursquare上，TGMF算法在准确率和召回率两个指标上均优于传统的兴趣点推荐算法。

关键词: 兴趣点推荐, 矩阵分解, 社交信任, BPR