Joint Geographical and Temporal Modeling based on Matrix Factorization for Point-of-Interest Recommendation

With the popularity of Location-based Social Networks, Point-of-Interest\n(POI) recommendation has become an important task, which learns the users'\npreferences and mobility patterns to recommend POIs. Previous studies show that\nincorporating contextual information such as geographical and temporal\ninfluences is necessary to improve POI recommendation by addressing the data\nsparsity problem. However, existing methods model the geographical influence\nbased on the physical distance between POIs and users, while ignoring the\ntemporal characteristics of such geographical influences. In this paper, we\nperform a study on the user mobility patterns where we find out that users'\ncheck-ins happen around several centers depending on their current temporal\nstate. Next, we propose a spatio-temporal activity-centers algorithm to model\nusers' behavior more accurately. Finally, we demonstrate the effectiveness of\nour proposed contextual model by incorporating it into the matrix factorization\nmodel under two different settings: i) static and ii) temporal. To show the\neffectiveness of our proposed method, which we refer to as STACP, we conduct\nexperiments on two well-known real-world datasets acquired from Gowalla and\nFoursquare LBSNs. Experimental results show that the STACP model achieves a\nstatistically significant performance improvement, compared to the\nstate-of-the-art techniques. Also, we demonstrate the effectiveness of\ncapturing geographical and temporal information for modeling users' activity\ncenters and the importance of modeling them jointly.\n