动态网络社团发现及演化模式分析方法  

Dynamic network community detection and evolution mode analysis method

作　　者：潘雨 姚锋[1] 刘鑫 张磊[3] 王帅辉 王沛[1] Pan Yu;Yao Feng;Liu Xin;Zhang Lei;Wang Shuaihui;Wang Pei(National University of Defense Technology,Changsha 410000,China;Army Engineering University,Nanjing 310000,China;Aca-demy of Military Science,Beijing 110000,China;Naval Command College,Nanjing 310000,China)

机构地区：[1]国防科技大学,长沙410000 [2]陆军工程大学,南京310000 [3]军事科学院,北京110000 [4]海军指挥学院,南京310000

出　　处：《计算机应用研究》2024年第12期3722-3728,共7页Application Research of Computers

基　　金：中国博士后科学基金资助项目(GZC20233530);国家社会科学基金资助项目。

摘　　要：如何充分考虑网络的演化过程准确发现动态网络的社团结构,并对社团演化模式进行跟踪和分析是动态网络社团发现的重要挑战。提出一种动态网络社团发现及演化模式分析算法EC-DCD。该算法利用前一时刻的社团发现结果作为先验信息来减少网络噪声对社团发现的影响,利用演化聚类框架平滑连续时刻的社团演化,获得每个时刻准确的社团结构。同时,引入社团演化矩阵对社团演化模式进行建模和跟踪,实现社团演化模式的分析和可视化。实验部分,将EC-DCD同基线算法FacetNet、DYNMOGA、DNMF、NE2NMF和CoDeDANet在人工数据集与真实数据集上进行了对比实验,实验结果证明EC-DCD不仅能够准确地划分每个时刻的社团结构,具有较强的稳定性,还能够跟踪社团的演化模式。How to obtain accurate community structure of the dynamic network,model the dynamic evolution process of the network,and realize the tracking and analysis of the community evolution mode is an important challenge for detecting dynamic network communities.This paper proposed a dynamic network community detection algorithm EC-DCD.The algorithm utilized the previous community detection results as a priori information to reduce the impact of network noise on community detection.It applied an evolutionary clustering framework to smooth the community evolution over consecutive time,achieving community structures at each time step.At the same time,it introduced the community evolution matrix to model and track the evolution mode of the community,which could realize the analysis and visualization of the community evolution mode.In the experiment,this paper tested the EC-DCD algorithm and baseline algorithms such as FacetNet,DYNMOGA,DNMF,NE2NMF,and CoDeDANet on the artificial datasets and the real datasets.The experimental results show that the proposed method EC-DCD can not only accurately detect the community structure at every moment,but also track the evolution pattern of the community.

关 键 词：社团发现 动态网络 演化聚类框架 非负矩阵分解 演化模式 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]