摘要
现有基于图自编码器的群体发现方法通常忽略了编码层多尺度特征对群体发现的影响,同时由于缺少统一的优化目标函数导致次优结果。为此,提出一种基于图自编码器多尺度特征融合的自监督群体发现方法。在图自编码器的基础上引入一种多尺度自表达模块,从不同编码层获取具有区分性的节点关系矩阵表示,并与节点潜在表示进行融合;通过节点聚类模块获得初步的群体识别结果;引入一种自监督模块监督节点表示学习过程,获得更优结果,构建一种端对端的网络群体发现模型。在多个公开数据集上进行对比实验,验证了所提方法的有效性,与现有方法相比,其在群体识别准确度上有了明显提升。
Existing methods based on graph autoencoder(GAE)usually ignore the influence of multi-scale feature information of encoding layers on community detection,and get sub-optimal results due to the lack of a unified optimization objective function.To this end,a self-supervised method for community detection based on GAE with multi-scale features was presented.A multi-scale self-expression module was introduced,discriminative node representations were obtained from different encoding layers and fused.A node clustering module was introduced to obtain rough clustering results.At the same time,a self-supervised node representation learning process was introduced to achieve better results,thereby constructing an end-to-end network group discovery model.Through comparative experiments on multiple datasets,the effectiveness of the proposed method is verified,and the accuracy of community detection is significantly improved compared with that of the existing methods.
作者
沈国栋
汪晓锋
毛岱波
王栽胜
张增杰
全大英
SHEN Guo-dong;WANG Xiao-feng;MAO Dai-bo;WANG Zai-sheng;ZHANG Zeng-jie;QUAN Da-ying(School of Information Engineering,China Jiliang University,Hangzhou 310000,China;Department of Information Technology,Ningbo Institute of Science and Technology Information and Development Strategy,Ningbo 315000,China)
出处
《计算机工程与设计》
北大核心
2024年第9期2805-2811,共7页
Computer Engineering and Design
基金
国家重点研发计划基金项目(2019YFB1707104)。
关键词
图自编码器
群体发现
多尺度特征
自监督学习
特征融合
端到端
统一优化
graph auto-encoder
community detection
multi-scale features
self-supervised learning
feature fusion
end-to-end
unified optimization
