Jointly Modeling Heterogeneous Temporal Properties in Location Recommendation

摘要

Point-Of-Interest (POI) recommendation systems suggest interesting locations to users based on their previous check-ins via location-based social networks (LBSNs). Individuals visiting a location are partially affected by many factors including social links, travel distance and the time. A growing line of research has been devoted to taking advantage of various effects to improve existing location recommendation methods. However, the temporal influence owns numerous dimensions which deserve to be explored more in depth. The subset property comprises a set of homogeneous slots such as an hour of the day, the day of the week, week of the month, month of the year, and so on. In addition, time has other attributes such as the recency which signifies the newly visited locations versus others. In this paper, we further study the role of time factor in recommendation models. Accordingly, we define a new problem to jointly model a pair of heterogeneous time-related effects (recency and the subset feature) in location recommendation. To address the challenges, we propose a generative model which computes the probability for the query user to visit a proposing location based on various homogeneous subset attributes. At the same time, the model calculates how likely the newly visited venues obtain a higher rank compared to others. The model finally performs POI recommendation through combining the effects learned from both homogeneous and heterogeneous temporal influences. Extensive experiments are conducted on two real-life datasets. The results show that our system gains a better effectiveness compared to other competitors in location recommendation.