软件导刊2025，Vol.24Issue(8)：49-58,10.DOI:10.11907/rjdk.241571
基于异构特征聚合的图神经网络算法设计
Design of Graph Neural Network Algorithm Based on Heterogeneous Feature Aggregation
摘要
Abstract
Heterogeneous network representation learning is important for exploring the features and properties of complex networks,thereby improving network security and the efficiency of social network analysis.However,existing research on heterogeneous networks tends to ignore the impact of network structure and its node heterogeneous content on the learning results,so this paper proposes a deep neural network algo-rithmic model HAGNN for heterogeneous feature aggregation.The method first extracts heterogeneous neighbouring nodes using the RWR mechanism;furthermore,Bi-GRU is used to learn complex structural information so as to perform feature aggregation on the neighbouring nodes of the sample;finally,the model was trained end-to-end using graph context loss method and small amount of gradient descent algo-rithm,and the attention mechanism was added for optimization.In the application experiment,the F1 value predicted by HAGNN in the node can reach 0.831,and the venue recommendation Recall value can reach 0.631,which is significantly improved compared with the relevant ap-plication results of other graph neural networks.关键词
异构网络表示学习/异构特征聚合/图神经网络Key words
heterogeneous network representation learning/heterogeneous feature aggregation/graph neural networks分类
信息技术与安全科学引用本文复制引用
侯浩民,陈昌奉,郭建邦..基于异构特征聚合的图神经网络算法设计[J].软件导刊,2025,24(8):49-58,10.基金项目
国家自然科学基金项目(62266019) （62266019）
湖南省自然科学基金项目(2024JJ7412) （2024JJ7412）
湖南省教育厅优秀青年项目(24B0496) （24B0496）
国家级大学生创新训练计划项目(202210531009) （202210531009）
