A Deep Point-of-Interest Recommendation System in Location-Based Social Networks.

Point-of-interest (POI) recommendation is an important part of recommendation systems in location-based social networks. Most existing POI recommendation systems, such as collaborative filtering based and context-aware methods, usually use hand-designed or manually selected features to achieve the recommendation. However, the information in the location-based social networks has very complicated relationships with each other, e.g., the latent relationships among users, POIs and user preferences, thus leading to poor recommendation accuracy. We propose a two-stage method to address this problem. In the first stage, user and POI profiles are abstracted using statistical methods. Then in the second stage, a deep neural network (DNN) is used to predict ratings on these candidate POIs, and finally the topN list of POIs is obtained. Experimental results on the Gowalla and Brightkite dataset show the effectiveness of our DNN based recommendation method.