Trust-aware spatial-temporal feature estimation for next POI recommendation in location-based social networks

Point of interest recommendation is one of the imperative tasks in location-based social networks. With the high influx of information, the recommendation has become a challenge. The collaborative filtering-based techniques have been plagued by implicit data sparsity and the presence of cold start users. To overcome such demerits, POI recommendation process must consider incorporating contextual information besides the user's check-in data. In this paper, we propose trust-aware spatial-temporal features for next POI recommendation model. To enhance the recommendation accuracy, we consider both the explicit and implicit trust of the users to decipher the POI preferences. The implicit trust is extrapolated from the user's check-in frequency, while the explicit trust is extracted based on the user's external social relations. We propose that the explicit social relations of a user encapsulate five levels of social connections: direct, transitive, temporal check-in based, location check-in based, and distance-confined social linkages. The two-phased process involves the user's neighborhood estimation to mine the propensity of the POIs for the users in the incipient phase and the neural collaborative filtering-based POI recommendation in the telic phase. The approach has been evaluated against two real-world datasets, namely Gowalla and Foursquare. The results juxtaposed with state-of-art approaches suggest the efficacy and importance of modeling social relations.