Scholarlink

<p>Graph neural networks (GNNs) have demonstrated remarkable success in classifying graph-structured data. However, conventional GNN architectures predominantly focus on node embedding representations derived from neighborhood characteristics, overlooking the crucial spatial topological information of nodes. To bridge this gap and enhance the discriminative capability of GNN, this paper establishes three graph neural networks with second-order pooling (SOP-GNN) anchor point selection, which are used for comparative studies of node classification models. Firstly, by leveraging second-order pooling, the model effectively identifies and designates important nodes as anchor points, capturing the hierarchical structure and significance within the graph data. Secondly, SOP-GNN constructs spatial topological and structural feature information of nodes by computing the global location relationships between captured nodes and anchor points. Finally, the model utilizes an attention mechanism to adaptively fuse the structural and topological features, generating comprehensive node embedding representations. Experiments on three real-world citation network datasets show that the SOP-GNN model achieves state-of-the-art performance in node classification tasks, validating its effectiveness in capturing spatial topological information and improving classification accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Node classification method for graph neural networks based on second-order pooling strategy

  • 期刊论文

  • 发表日期：

    2026年7月2日

查看全文
  • Xiwen Qin,
  • Xueteng Cui,
  • Yu Jia,
  • Dingxin Xu

摘要

Graph neural networks (GNNs) have demonstrated remarkable success in classifying graph-structured data. However, conventional GNN architectures predominantly focus on node embedding representations derived from neighborhood characteristics, overlooking the crucial spatial topological information of nodes. To bridge this gap and enhance the discriminative capability of GNN, this paper establishes three graph neural networks with second-order pooling (SOP-GNN) anchor point selection, which are used for comparative studies of node classification models. Firstly, by leveraging second-order pooling, the model effectively identifies and designates important nodes as anchor points, capturing the hierarchical structure and significance within the graph data. Secondly, SOP-GNN constructs spatial topological and structural feature information of nodes by computing the global location relationships between captured nodes and anchor points. Finally, the model utilizes an attention mechanism to adaptively fuse the structural and topological features, generating comprehensive node embedding representations. Experiments on three real-world citation network datasets show that the SOP-GNN model achieves state-of-the-art performance in node classification tasks, validating its effectiveness in capturing spatial topological information and improving classification accuracy.