计算机应用研究2026，Vol.43Issue(3)：908-916,9.DOI:10.19734/j.issn.1001-3695.2025.06.0267
基于自注意力机制的全局社区隐藏算法
Global community hiding algorithm based on self-attention mechanism
摘要
Abstract
Community detection algorithms demonstrate strong capability in mining network data but simultaneously pose risks of user information leakage.To address this issue,researchers actively explore community hiding as a solution.However,most existing studies on community hiding focus on topological networks and achieve limited progress in attributed networks.To over-come this limitation,this study proposed a self-attention mechanism network community hiding algorithm.The algorithm mo-deled the global dependencies among node attributes using a self-attention mechanism and generated adversarial perturbations by integrating structural information of the graph.It computed attention weights between nodes to derive node embeddings and then estimated the probability of link existence.Based on a predefined perturbation budget,the algorithm selected links with the greatest impact on the community structure for addition or deletion,thereby achieving effective community hiding.Experimental results demonstrate that SANCH delivers outstanding hiding performance and robustness against various community detection algorithms.关键词
图神经网络/自注意力机制/属性网络/社区隐藏/隐私保护Key words
graph neural network/self-attention mechanism/attributed network/community hiding/privacy protection分类
信息技术与安全科学引用本文复制引用
冯志超,张博瀚,荆军昌,李永波,刘栋..基于自注意力机制的全局社区隐藏算法[J].计算机应用研究,2026,43(3):908-916,9.基金项目
国家自然科学基金资助项目(62072160) （62072160）
河南省科技攻关计划资助项目(242102211076) （242102211076）
