基于改进的标签传播算法的网络聚类方法  

Network clustering method based on improved label propagation algorithm

作　　者：桂春[1] 黄旺星 GUI Chun1 , HUANG Wang-xing2(1. School of Mathematics and Computer Science, Northwest University for Nationalities, Lanzhou 730030, China; 2. School of Electronic Information, Wuhan University, Wuhan 430072,China)

机构地区：[1]西北民族大学数学与计算机科学学院,兰州730030 [2]武汉大学电子信息学院,武汉430072

出　　处：《吉林大学学报（工学版）》2018年第5期1600-1605,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：甘肃省地区科学基金项目(61562075);西北民族大学甘肃省一流学科专项经费项目;中央高校基本科研业务费专项项目(31920180119)

摘　　要：采用传统标签传播算法实现网络聚类时,由于标签初始分配过程随机、节点选择过程随机、且标签更新顺序随机的原因,影响聚类结果。为此,提出一种新的基于改进标签传播算法的网络聚类方法,即用图对网络进行描述,并为网络聚类提供基础。改进标签传播算法过程如下:求出网络中任意两节点拥有最大公共邻居的平均阶数,把相似性最高的节点和邻居节点看作初始核心社团,为其分配初始标签;引入基于随机游走的相似度矩阵,令节点选择和自身相似度最高的节点拥有的标签;通过H指数对标签算法更新顺序进行改进;依据改进后结果,按照标签传播算法网络聚类过程实现聚类。实验结果表明,本文所提的网络聚类方法具有更高的准确性和稳定性。Using traditional label propagation algorithm in network clustering,the initial allocation of taps,the node selection and updating order of the labels are all random processes,which affect the clustering results.To overcome this problem,a new method of network clustering based on improved label propagation algorithm is proposed,in which the graph is used to describe the network and provide the basis for the network clustering.The process of the improved label propagation algorithm includes the following steps.First,the average malpractice in the network order of any two nodes with the largest common neighbor number is obtained.Second,the most similar node and neighbor node are taken as the initial core community and its initial label is assigned.Third,the random walk similarity matrix is introduced based on the node selection and the label of the node with the highest similarity.Fourth,by using the H index on the label algorithm,the update order is improved.Finally,according to the improved result,the clustering process is realized according to the clustering process of the label propagation algorithm network.Experimental results show that the proposed network clustering method has higher accuracy and stability.

关 键 词：计算机应用 节点 标签 传播算法 网络聚类 相似度 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]