动态复杂网络社区挖掘—选择性聚类融合算法

摘要

Research on dynamic complex network mining is a hot topic currently. Based on selective clustering fusion, this paper proposes a dynamic complex network mining algorithm. Firstly, the algorithm divides the dynamic process into snapshots with same time interval. According to the conceptions such as similarity and centrality of vertices, every snapshot gets the corresponding clustering outcomes in a speed accelerated by an improved hierarchical clustering algorithm. Secondly, the clustering results collection have to be selected on the basis of difference between clustering outcomes with aim to get various clustering members required in the following fusion process. Lastly, this paper comes up with the conception of weighted Co-association matrix in terms of time attenuation, and then obtains the final clustering results using the single-link algorithm. The clustering accuracy and the degree of dynamic characteristic mining are tested in the stochastic network and the real network. Experimental results demonstrate the feasibility and validity of this algorithm.%针对当前研究动态复杂网络的热点问题,提出了一种基于选择性聚类融合的社区挖掘算法.该算法首先将动态过程划分为相同时间间隔的快照,利用欧几里德距离、顶点权重等技术,使用一种改进的层次聚类算法加快聚类速度,得到每个快照相应的聚类结果；然后根据这些聚类结果之间的差异性,筛选聚类结果集合,为融合过程提供多样性的聚类成员；考虑到时间衰减性,设计了加权共联矩阵,使用单链接算法来得到最终的聚类结果.在随机网络和真实世界网络上对算法的计算速度和动态特征挖掘情况两方面进行实验,结果表明了该算法的可行性和有效性.