Mining dynamic preferences from geographical and interactive correlations for next POI recommendation

Next point-of-interest recommendation has become an increasingly significant requirement in location-based social networks. Recently, RNN-based methods have shown promising advantages in next POI recommendation due to their superior abilities in modeling sequential transitions of user behaviors. Despite their success, however, exploring complex correlations between POIs and capturing user dynamic preferences are still challenging issues. To overcome the limitations, we propose a novel framework named MPGI (Mining Preferences from Geographical and Interactive Correlations) for next POI recommendation. Specifically, we first design a POI correlation modeling layer to capture geographical distances and interactive correlations between all of POI pairs. Then, we fuse relevant signals from highly correlated POIs into target POI for high-quality POI representations. Furthermore, for user long- and short-term preferences modeling, we propose position-aware attention unites and attention network to dynamically select the most valuable information in check-in trajectories. Experimental results on two real-world datasets demonstrate that MPGI consistently outperforms the state-of-the-art methods.

Automated summary

This article introduces a novel framework named MPGI for next POI recommendation. The framework captures geographical and interactive correlations between POIs and dynamically selects user preferences, improving the quality of recommendations.