Deriving an effective hypergraph model for point of interest recommendation

Point of interest (POI) recommendation on Location Based Social Networks (LBSN) is challenging as the data available for predicting the next point of interest is highly sparse. Addressing the sparsity issue becomes one of the keys to achieve accurate POI recommendation. A promising approach is to explore various types of relevant information carried by the network, e.g, network structures, spatial-temporal information and relations. In this paper, we put forward a hypergraph model to incorporate the higher-order relations of LBSNs for POI recommendation. Accordingly, we propose a hypergraph random walk (HRW) to be applied to such a complex hypergraph. The steady state distribution gives our derived recommendation on venues for each user. Experiments based on a real data set collected from Foursquare have been conducted to evaluate the efficiency and effectiveness of our proposed model with promising results obtained.