POI Recommendation Based on Heterogeneous Network

With the development of wireless networks and positioning technologies, location-based social networks (LBSN) have gained popularity. More and more people share experiences about points of interest (POI) through “check-in” behavior. Mining the check-in data can help people discover the POI they are interested in. However, the data sparsity of user check-in records and the cold start problem with users and POI pose serious challenges. In addition, POI recommendation need to consider the impact of multiple factors. In order to solve the above problems, we propose a POI recommendation method based on heterogeneous network representation learning, called HRPR. First, we propose to use the meta-path based weighted random walk method to generate node sequences and learn the representation vector of the user and POI by means of the skip-gram model. Then, we design a POI recommendation framework based on deep neural network. The experimental results on real-world Yelp dataset demonstrate the effectiveness of our framework.