Spatio-Temporal Unequal Interval Correlation-Aware Self-Attention Network for Next POI Recommendation

Abstract

As the core of location-based social networks (LBSNs), the main task of next point-of-interest (POI) recommendation is to predict the next possible POI through the context information from users’ historical check-in trajectories. It is well known that spatial–temporal contextual information plays an important role in analyzing users check-in behaviors. Moreover, the information between POIs provides a non-trivial correlation for modeling users visiting preferences. Unfortunately, the impact of such correlation information and the spatio–temporal unequal interval information between POIs on user selection of next POI, is rarely considered. Therefore, we propose a spatio-temporal unequal interval correlation-aware self-attention network (STUIC-SAN) model for next POI recommendation. Specifically, we first use the linear regression method to obtain the spatio-temporal unequal interval correlation between any two POIs from users’ check-in sequences. Sequentially, we design a spatio-temporal unequal interval correlation-aware self-attention mechanism, which is able to comprehensively capture users’ personalized spatio-temporal unequal interval correlation preferences by incorporating multiple factors, including POIs information, spatio-temporal unequal interval correlation information between POIs, and the absolute positional information of corresponding POIs. On this basis, we perform next POI recommendation. Finally, we conduct comprehensive performance evaluation using large-scale real-world datasets from two popular location-based social networks, namely, Foursquare and Gowalla. Experimental results on two datasets indicate that the proposed STUIC-SAN outperformed the state-of-the-art next POI recommendation approaches regarding two commonly used evaluation metrics.