基于边聚类系数的谱聚类社区划分方法研究  

Spectral clustering community partition method based on edge clustering coefficient

作　　者：赵菲 余本国[1] 冀庆斌[1] ZHAO Fei;YU Benguo;JI Qingbin(School of Science,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学理学院

出　　处：《华中师范大学学报（自然科学版）》2020年第1期17-22,共6页Journal of Central China Normal University：Natural Sciences

基　　金：国家自然科学青年基金项目(61807005);山西省教改(研究生)项目(2017JG74)

摘　　要：针对图谱划分方法在划分社区结构不是很明显的网络时,不能得到好的划分效果,该文提出了基于边聚类系数的谱聚类社区划分方法.由于社区内部节点之间的连接比各个社区间节点的连接稠密,边聚类系数的大小反映了节点的聚集程度,因而通过网络中的边所构三角形的数量定义了聚类系数矩阵,矩阵中的元素即处于网络中的边实际构成三角形的数量.在增益函数最大化的过程中,使用了矩阵的特征值和特征向量,以此来进行社区划分.通过在真实网络数据中进行实验,结果表明该算法可行.As for the graph partition method, it can not get good partition effect when the network of community structure is not obvious, a spectral clustering community partition method based on edge clustering coefficient is proposed. Because the connection between nodes in the community is denser than that between nodes in each community, the size of edge clustering coefficient reflects the aggregation degree of nodes. Therefore, the clustering coefficient matrix is defined by the number of triangles constructed by edges in the community, and the elements in the matrix are the numbers of triangles actually formed by edges in the community. In the process of maximizing the aggregation degree function, the eigenvalues and eigenvectors of the matrix are used to divide the community. Experiments on real network data show that the algorithm is feasible.

关 键 词：复杂网络 边聚类系数 聚类系数矩阵 增益函数 社区划分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]