Exploring Sequential and Collaborative Contexts for Next Point-of-Interest Recommendation

Sequential methods based on recurrent neural networks, as well as session-based long- and short-term approaches have become the state-of-the-art methods for next Point-of-Interest (POI) recommendation recently. However, most of them use spatial and temporal correlations among check-ins while they fail to model category-based sequences explicitly. Moreover, most of the session-based methods only consider users’ own sessions, while they neglect the information from collaborative sessions. Besides, most of the sequential methods only consider the information in users’ own sequences, while they neglect inherent similarities among POIs from a global perspective. To this end, we propose a method to explore sequential and collaborative contexts (SCC) for next POI recommendation. We simultaneously model temporal, spatial and categorical correlations among check-ins to capture sequential contexts. We generate collaborative sessions for current sessions, then leverage the collaborative information to better predict users’ recent visit intents. Besides, a similarity graph is proposed to leverage collaborative information from POI side on a global scale. Finally, we combine sequential and collaborative contexts to capture preferences of users. Extensive experimental results demonstrate our model outperforms other state-of-the-art baselines consistently.