基于传播影响力的重叠社区划分算法

Overlapping Community Detection Algorithm According to Propagating Influence

在研究经典标签传播算法的基础上,提出了一种基于传播影响力的重叠社区划分算法COPRA-PI,可用于挖掘加权网络中的社区结构。该算法在COPRA算法的基础上从节点影响力、边影响力、历史标签影响力3个方面综合考虑传播影响力;同时针对COPRA算法中每个节点在每次迭代过程中均具有相同的最大标签数,且该最大标签数目需手动设置等不足,该算法中设计了一个自适应的最大标签数。实验结果表明,COPRA-PI算法在经典的数据集上对比现有经典算法更能挖掘出高质量的社区结构且收敛速度较快。

This paper proposed an overlapping community detection algorithm according to propagating influence, called COPRA-PI,to mine community structure in weighted networks. COPRA-PI algorithm considers the influence of propagation from three aspects based on conventional COPRA： node influence, edge influence and historical label influence. For addressing the issues of that each node in COPRA has the same maximum number of label which needs to be set manually at each moment, an adaptive maximum number of label is also designed. Experimental results show that the proposed COPRA-PI not only can detect the community division accurately but also has a fast convergence rate.