基于节点向量和密度峰值的重叠社团检测方法

Overlapping community detection method based on node vector and density peak

在真实世界网络中,数据量往往较大、维度较高,这使得数据难以处理,并且所包含的社团大多重叠,而大多数已经存在的算法针对的是非重叠社团,基于改进的密度峰值和标签传播的重叠社团检测算法(DPCL算法).采用低维向量表示网络中节点信息,根据节点的局部密度和相对距离选出中心节点.将只与一个中心节点直接相邻的节点分配到该中心节点所在的社团内,对剩余节点通过计算归属度进行分配,从而实现对重叠社团进行检测;在真实世界网络和LFR基准合成网络上与其他社团检测方法进行比较,实验结果表明能够有效的检测重叠社团.

In real-world networks,the amount of data was often large and the dimension was high,which makes the data difficult to process,and most of the communities contained were overlapped,while most existing algorithms were aimed at non overlapped communities.The overlapped community detection algorithm based on improved density peak and label propagation(DPCL algorithm)first used low-dimensional vectors to represent the node information in the network,the central node was selected according to the local density and relative distance of the node;the nodes directly adjacent to one central node were allocated to the community where the central node was located,and the remaining nodes were allocated by calculating the attribution degree,so as to detect the overlapping community,compared with other community detection methods on real-world network and LFR benchmark synthetic network,the experimental results showed that it can effectively detected overlapping communities.