Collaborative tensor factorization and its application in POI recommendation

The rapid prevalence of location-based services provides an opportunity to understand people's mobility behavior, helps people easily decide where to go at the next moment, and helps salesmen select the most suitable commercial sites to delivery advertisements. In this paper, we propose a POI recommendation method based on collaborative tensor factorization. Firstly, we present a generalized objective function for collaboratively factorizing an n-mode tensor with m feature matrices. Secondly, all users' check-in behaviors are modeled as a 3-mode tensor and three feature matrices are extracted to characterize the time distribution, category distribution and POI correlation. Thirdly, we fill the missing entries by using our collaborative tensor factorization approach such that a user's rating to a POI at a specific time is recovered. Our experiment on a real check-in database shows that our method can provide more accurate recommendation.