福州大学学报（自然科学版）2025，Vol.53Issue(2)：127-134,8.DOI:10.7631/issn.1000-2243.24156
融合拉普拉斯位置编码和自注意力机制的图神经网络
Graph neural network incorporating Laplacian position encoding and self-attention mechanism
摘要
Abstract
In order to solve the problem of insufficient structure information captured by existing posi-tion encoding and limited expression ability of spectral graph filters,the LESpecformer model architec-ture based on Laplacian position encoding and self-attention mechanism is proposed.Firstly,this paper introduces the Laplacian position encoding for sensing the position information of different nodes in the graph structure,which improves the ability of the model to learn the relative position information of different nodes,and then captures the global information of the graph structure.Secondly,based on the set of eigenvalues fused with Laplacian position encoding,the self-attention mechanism is used to adaptively learn the dependencies between eigenvalues and obtain effective new basis representations,which facilitates the proposed model in learning better node embeddings,and thus improves the accu-racy of node classification.Finally,compared with different baseline networks on six graph datasets,the experimental results show that the performance of the proposed LESpecformer is optimal.关键词
谱图滤波器/拉普拉斯位置编码/自注意力机制/图神经网络Key words
spectral graph filters/Laplacian position encoding/self-attention mechanism/graph neural network分类
信息技术与安全科学引用本文复制引用
邹成龙,李伟诺,黄梅香,林艺东..融合拉普拉斯位置编码和自注意力机制的图神经网络[J].福州大学学报（自然科学版）,2025,53(2):127-134,8.基金项目
国家自然科学基金资助项目(12201284) （12201284）
福建省自然科学基金资助项目(2022J05169) （2022J05169）
闽南师范大学基金资助项目(KJ2021020&MSGJB2022010) （KJ2021020&MSGJB2022010）
