一种全局图增强的图神经网络新闻推荐算法

Graph Neural Network News Recommendation AlgorithmBased on Global Graph Enhanced

针对现有图神经网络新闻推荐方法用户兴趣建模角度单一,且无法迅速拟合新节点特征的问题,提出一种全局图增强的图注意力网络模型,在以全局图采样子图的方式聚合邻居节点特征的同时,综合考虑用户历史时序特征和类别特征,多层级地建模用户兴趣。在MIND数据集上通过大量实验表明,提出的模型优于现有的基线网络方法。

To address the limitations of existing graph neural network-based news recommendation methods,which often suffer from a simplistic modeling of user interests and the inability to rapidly adapt to new node features,a novel global graph-enhanced graph attention network(GGE-GAT)model was proposed in this study.By aggregating neighbor node features using subgraph sampling from a global graph,the proposed model comprehensively models user interests by considering both user historical temporal features and category features in a multi-level manner.Extensive experimentation on the MIND dataset demonstrates the superiority of the proposed model over existing baseline network methods.