STIN: Spatio-Temporal Inference Network Enhanced by Social Attributes for Next Location Prediction

The next location prediction holds a pivotal role within the domain of location-based social network applications. It aims to predict the location where users are most likely to visit in the next time step based on their historical trajectories. However, traditional methods only modeled the spatio-temporal data from the dimension of temporal and spatial attributes of the trajectory, ignoring the hidden social information which is also essential in this task. This makes it difficult for the model to effectively use the information of trajectory period changes of different lengths and to predict accurately in the face of sudden trajectory changes. To solve this problem, this paper proposes a Spatio-Temporal Inference Network (STIN) incorporating the social attributes with spatio-temporal information that enhances the conventional user trajectory representation. We describe the social attributes into two different types: explicit social attributes among the user periodic behavior and hidden social attributes among various categories of geographical locations. By subdividing the time dimension into months and weeks, STIN incorporates additional periodic temporal features to explore more social patterns. Additionally, STIN introduces external knowledge to the model and describes the hidden social attributes among the locations of user trajectories. Experimental results show that our method has effectively improved on three commonly used real-world LBSN datasets: Foursquare NYC, Foursquare TKY, and Gowalla. The performance of STIN on these datasets is superior to that of the existing state-of-the-art methods, with improvements of 9.78%, 4.99%, and 21.87% in MRR metrics.