一种基于评分矩阵局部低秩假设融合地理和文本信息的协同排名POI推荐模型*

针对目前LBSN中，用户只对少数兴趣点进行签到，使得用户签到历史数据及其上下文信息（如评论文本）极其稀疏，同时传统的评分推荐系统只考虑用户和评分二元信息，具有一定的局限性。为此，提出一种基于评分矩阵局部低秩假设的局部协同排名兴趣点推荐算法。首先，假设用户-兴趣点矩阵在由用户-兴趣点对所定义度量空间中某些邻域内是低秩；其次，对于地理信息建模采用一种自适应二维核密度方法，然后，对于文本信息利用潜在狄利克雷分配模型挖掘兴趣点相关的文本信息建模用户的兴趣主题；最后，基于局部协同排名模型将兴趣点的地理信息和评论文本信息有效融合。实验结果表明:该模型的性能优于主流先进兴趣点推荐算法。

A synthetic rank-oriented recommendation model for point-of-interest recommendation algorithm based on the assumption of locally low-rank rating matrix: exploiting review information and geographical information

Since the user only check-in a few POIs in LBSN, so that the historical data of users and its context information (such as review information, geographic information and so on) are extremely sparse, while the traditional recommendation system only consider the user and score binary values, which create a severe challenge. To cope with this challenge, this paper proposed to utilize local collaborative ranking (LCR) based on the assumption of locally low-rank rating matrix model for POI recommendation. Firstly, this paper assumed that the rating matrix is low-rank within certain neighborhoods of the metric space defined by (user, POI) pairs, which assumed the user-POI matrix is locally low-rank instead of globally low-rank. Second, this paper modeled a personalized check-in probability density with an adaptive bandwidth over the two-dimensional geographic coordinates for each user. Third, this paper exploited an aggregated Latent Dirichlet Allocation (LDA) model to learn the interest topics of users and inferred the interest POIs by mining textual information associated with POIs and generate interest relevance score. Further, this paper exploited probabilistic matrix factorization model (PMF) to integrate the review and geographical for POI recommendation to increase the density of local matrix and improved the accuracy of recommendation. Experimental results show that RG-LCR outperform other state-of-the-art POI recommendation algorithms.