Beyond fixed time and space: next POI recommendation via multi-grained context and correlation

POI recommendation is significant for discovering attractive locations, crime prediction, and smart city construction. Most existing methods only consider the fixed time and space between successive check-in points when capturing sequential patterns from trajectory history. However, single granularity is inadequate to mine the spatial-temporal influence on sequential patterns in sparse and incomplete check-in data. Besides, they neglect the relevance between non-adjacent check-ins and fail to fully exploit factors for the correlation mining. To tackle the above issues, we propose a novel model for the next POI recommendation via multi-granularity context and correlation. It focuses on exploring vital factors for modeling effective spatial-temporal contexts and mining potential correlations among check-ins. Specifically, for context modeling, we explore effective spatial-temporal contexts to learn mobility patterns locally and globally by introducing hierarchical regions and slots. For correlation modeling, we only focus on the geographical influence. We employ a spatial-aware function to measure the correlations among check-ins to find the predictive ones for the recommendation. Extensive experiments on widely used datasets indicate that our MGCOCO consistently and significantly outperforms the state-of-the-art approaches.