Muti-channel graph attention networks for POI recommendation

This paper focuses on the task of Point-of-interest (POI) recommendation whose goal is to generate a list POIs for target user based his or her history check-in records. Different from traditional tasks (e.g., movie recommendation), there are many factors, like temporal factor and geographical factor, which make great influence preference. Though existing POI methods tend model preference social they fail these factors into jointly model, leading learn suboptimal To tackle this issue, we propose Muti-channel Graph Attention Network (MGAN) learns multiple aspects in unify model. Specifically, MGAN first constructs several graphs with corresponding contextual features capture temporal, geographical, semantic aspects. Then leverages graph attention networks representations graphs. Finally, estimates records other similar users via learned representations. We conduct extensive experiments real-world datasets. And results indicate that our proposed outperforms mainstream methods.