Personalized location recommendation for location-based social networks

With the development of social networks and wireless communication technology, location-based social networks (LBSNs) are developing rapidly. Personalized location service in location-based social networks can provide users with a new point-of-interest (POI). Compared to traditional recommendation, point-of-interest recommendation integrates the social network with the location that connects online users and physical places. In current studies, user preference, social influence and geographical influence are mostly taken into consideration to recommend satisfying point-of-interests to users. Although geographical context is significant, most articles only consider spatial properties of geographical influence. In addition to spatial properties of geographical influence, we put a special emphasis on sequence properties of that to exploit implicit dependencies between POIs. We model spatial properties using a kernel density estimation approach to generate a unique distribution for each user. Furthermore, user preference, social influence, and sequence properties of geographical influence are integrated into a random walk model on graph. Finally, we conduct a comprehensive performance evaluation over two real-world datasets collected from Brightkite and Gowalla. Experimental results show that the proposed recommendation approach performs better than other methods.