Existing point-of-interesting (POI) recommendation methods lack sufficient integration of information related to the features of individual users and their corresponding contexts. Personalized location-based services built upon these methods have a few limitations such as low recommendation accuracy and untapped potential interests of users. To overcome these limitations, we propose an efficient POI recommendation method based on multiple factors (i.e., preference, social relationship, and spatial–temporal factors) to improve the recommendation accuracy and alleviate the cold-start and sparsity problems. Unlike existing methods, the proposed method considers the features of each factor. First, we identify direct and indirect trust relationships, upon which the improved trust relationship measurement methods are built respectively. Second, we fuse the comprehensive trust relationship, user preference, check-in time, and geographical location into a matrix factorization model. We pay special attention to the attraction of items, correlation between items, trajectory composed of locations, and the influence of interest-forgetting during the fusing process. Finally, we generate recommendation lists of POIs for the target users. With three real-world datasets, experimental and analytical results show that the proposed method outperforms existing methods, while alleviating the cold-start and sparsity problems that commonly hinder POI recommender systems. Theoretically, our study contributes to the effective usage of multidimensional data science and analytics for POI recommender system design. In practice, our results can be used to improve the quality of personalized POI recommendation services for websites and applications.

现有的兴趣点 (POI) 推荐方法缺乏与个人用户的特征及其相应上下文相关的信息的充分整合。基于这些方法构建的个性化基于位置的服务具有一些限制，例如推荐准确度低和用户的潜在兴趣未开发。为了克服这些限制，我们提出了一种基于多种因素（即偏好、社会关系和时空因素）的高效 POI 推荐方法，以提高推荐精度并缓解冷启动和稀疏问题。与现有方法不同，所提出的方法考虑了每个因素的特征。首先，我们确定了直接和间接信任关系，在此基础上分别建立了改进的信任关系度量方法。第二，我们将综合信任关系、用户偏好、签到时间和地理位置融合到一个矩阵分解模型中。我们特别注意项目的吸引力，项目之间的相关性，由位置组成的轨迹以及在融合过程中遗忘兴趣的影响。最后，我们为目标用户生成 POI 的推荐列表。通过三个真实世界的数据集，实验和分析结果表明，所提出的方法优于现有方法，同时缓解了通常阻碍 POI 推荐系统的冷启动和稀疏问题。从理论上讲，我们的研究有助于将多维数据科学和分析有效地用于 POI 推荐系统设计。在实践中，