动态网络保留社区结构嵌入  

Dynamic network embedding of preserved community structure information

作　　者：彭姿文 周建国[1] PENG Ziwen;ZHOU Jianguo(School of Electronic Information,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学电子信息学院,湖北武汉430072

出　　处：《武汉大学学报(工学版)》2025年第2期325-333,共9页Engineering Journal of Wuhan University

摘　　要：针对现有的网络嵌入方法大多只考虑静态网络,应用于动态网络时无法保留时间相关性,以及现有的动态网络嵌入方法仅保留网络的局部结构信息而忽略网络全局结构的问题,考虑到属于同社区的节点在低维空间也应更接近,提出一种动态网络保留社区结构(dynamic network preservation community structure,DynPCS)嵌入方法。设计模型以协同学习网络嵌入和社区检测,将网络的局部和全局特征转化为低维稠密的实数向量,并利用长短期记忆网络学习网络的动态变化,捕获动态网络的演化模式。在4个真实数据集上进行实验,结果表明:所提方法应用于下游社区检测任务时,在归一化互信息和调整兰德系数2个指标上均优于现有的相关方法,能够保留网络社区结构,挖掘网络潜在的隐藏关系。Most of the current network embedding approaches merely consider static networks and fail to retain temporal correlation when applied to dynamic networks.Moreover,the existing dynamic network embedding methods merely preserve the local structural information of the network while neglecting the global structure of the network.Considering that nodes belonging to the same community should also be closer in low-dimensional space,this paper proposes a dynamic network embedding method(DynPCS)that can preserve the community structure.The model designed collaboratively learns network embedding and community detection,the local and global features of the network are transformed into a low-dimensional dense real number vector,and the dynamic changes are learned by using a long and short-term memory network to capture the evolutionary patterns.Experiments on four real datasets show that the proposed method outperforms the existing related methods in both normalizing mutual information(NMI)and adjusting RAND index(ARI)when applied to the downstream community detection task,and can preserve the network community structure and uncover the potential hidden relationships in the network.

关 键 词：网络表示学习 动态网络 社区检测 概率生成模型 长短期记忆网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]