MGTIPM: A Multi-granularity Time Intention Prediction Model for Next Point-of-Interest Recommendation

Next Point-of-Interest (POI) recommendation aims at ranking several POIs that the user tends to visit next time based on the user's historical trajectories. Most of the previous studies directly predict the next check-in POIs. However, the user's time intention and revisit preference play important roles in the effective recommendation and can be excavated based on the strong relation between check-in time and location. Therefore, in this study, a new model named MGTIPM is discussed. Specifically, MGTIPM designs a multi-granularity time intention prediction module to effectively predict the next check-in time within a period of one day. Moreover, a balancing mechanism integrates the time intention into the check-in sequence POI recommendation model so that MGTIPM can recognize the users' intention to revisit POIs. Experiments are conducted on two datasets and the results show that the proposed MGTIPM outperforms baselines regarding recommendation performance.