基于密度峰值和网络嵌入的重叠社区发现

Overlapping community detection based on density peaks and network embedding

密度峰值是一种基于密度的聚类算法,该算法假设类簇中心点具有较高的密度且被密度较小的节点包围。由于图结构的性质,密度峰值无法直接适用于网络结构,现有的基于密度峰值的社区发现算法大部分是基于图的拓扑结构或者邻接矩阵度量节点近似度,这种方法往往引入较大的计算复杂度。文中结合网络嵌入方法通过低维向量表示网络中的节点信息,提出了一种基于密度峰值和网络嵌入的重叠社区发现算法(overlapping community detection based on density network embedding, OCDDNE)。该算法首先通过网络嵌入获取节点的网络结构特征,然后基于改进的密度峰值的方法对嵌入后的节点向量进行多标签聚类,使编码后的向量之间的结构关系得到更好的揭示,从而发现网络中的重叠社区结构。在人工网络和真实网络的验证实验表明,该算法可以有效的挖掘网络中的重叠社区结构,并在结构复杂度较高的网络中优于其他算法。

Density peaks is a density-based clustering algorithm.It assumes that the center points of the cluster have higher density and are surrounded by nodes with lower density.Due to the character of graph structure,density peaks cannot be directly applied to network structure,and most of density peaks based community detection algorithms based on density peaks are based on graph topology or adjacency matrix to measure node approximation,which often leads to great computational complexity.This paper proposes an overlapping community detection algorithm based on density and network embedding(OCDDNE).The proposed algorithm firstly embeds the network structure characteristics of nodes in network structure through network embedding,and then clusters the embedded node vectors based on the improved method of density peaks,so that the structural relationship between the encoded vectors can be better revealed and the overlapping communities which each node is located is determined.Experiments on synthetic networks and real networks data show that the proposed algorithm can efficiently find the overlapping community structure in networks,and it is superior to other algorithms especially in complex networks with high structural complexity.