A Geographical Behavior-Based Point-of-Interest Recommendation

With the development of mobile devices, point-of-interest (POI) recommendation has received increasing attention. However, achieving accurate personalized POI recommendation is challenging due to the sparsity of the available data per user. In addition, previous efforts based on collaborative filtering mainly treat user behavior as a whole part in computing user similarity which sometimes cannot yield inferior performance. In this paper, we propose a novel two-phase algorithm to boost personalized POI recommendation performance by incorporating three unique characteristics in Location-Based Social Networks (LBSNs), namely, activity-based periodic behavior, time-aware multi-centers geographical behavior, and spatio-temporal relation. Experiments show that our proposed approach can further improve the recommendation accuracy and efficiency compared to previous works.