自适应的邻接关系:去相关图方法在下个兴趣点推荐的应用OA

The task of next Point-of-Interest(POI)recommendation plays a significant role in social networking services.Its goal is to predict the next POI that a user is most likely to visit,based on their historical check-in records.Existing methods generally construct relational graphs for POIs using predefined graphs,while only a few studies have explored adaptive graph approaches.Adaptive graph representation learning can capture more meaningful graph structures,enabling the subsequent propagation process in Graph Neural Networks(GNNs)to learn more significant adjacency relationships,and thus obtain more valuable POI embedding.This enables downstream sequence modeling tasks to capture potential relationships between POIs better.However,research on adaptive graphs remains in its early stages.In this paper,we propose a Decorrelated Graph Representation-enhanced Attention Network(DGRAN)for the next POI recommendation task.Additionally,this paper explores the relationship between the self-attention mechanism and adaptive graph learning,introducing extra residual connections to self-attention methods in this field to increase the gradient and ensure the hight quality updates of the adaptive graph structure.Results on two real-world datasets demonstrate that the method of this paper outperforms state-of-the-art baselines.