Leveraging GCN and User Friendship for Next POI Recommendation

The next POI (point of interest) recommendation aims to explore the behaviour patterns from users’ historical check-in records and recommend the next location. However, solving the data sparsity problem and improving the model recommendation performance are still considerable challenges in POI recommendations. This paper proposes a GFUC model consisting of the GCN (Graph Convolutional Networks) and FUC (Friendship of User Context) modules. The GCN module integrates weighted graph convolutional networks to obtain the best representation of users and POIs. The FUC module divides user check-in records with multiple fine-grain manners. Then it incorporates the user friendship and context information, significantly improving the model’s performance while solving the data sparsity problem. More specifically, experimental results show that our proposed GFUC model improves the performance of POI recommendations by more than 10% on both Yelp and Gowalla datasets with the evaluation metrics Precision@10.