Graph-Enhanced Spatio-Temporal Interval Aware Network for Next POI Recommendation in Mobile Environment

Abstract

With the rapid spread of mobile device, technologies in mobile cloud increased quick and introduce huge volume of mobile data and computation. Human movement between POIs are recorded in mobile cloud, which indicate personalized behaviors. Most POI recommendation method in mobile cloud proposed to utilize recurrent neural networks and self-attention mechanism to discover users’ potential movement behaviors. In this paper, we propose a graph-enhanced spatio-temporal interval aware network (GESTIAN) to recommend the next POI. In GESTIAN, we propose a graph-based general pattern learning module to learn common behavior patterns based on a global trajectory flow graph to address the challenges caused by cold start. Furthermore, we propose a heterogeneous network with spatio-temporal interval aware with self-attention and gate recurrent unit to extract users’ long-term and short-term spatio-temporal dependencies, respectively. In addition, we leverage the scale between positive and negative samples by randomly sampling negative samples. We conduct extensive experiments based on two real-world check-in datasets. The experimental evaluations demonstrate that the proposed GESTIAN outperforms most challenging baselines by approximately 2%-10%, and achieves better performance over few-check-in history users.