复杂网络社团挖掘中基于路径的模块性分析方法

Path-based Modularity Analysis Method for Community Detection of Complex Network

在复网络社团挖掘过程中,传统的模块性定义依赖网络的邻接矩阵,仅考虑相邻节点间的关系,不能很好地描述真实网络的结构特征。为此,提出一种基于路径的模块性分析方法。根据网络节点间的路径及其概率分布构建模块性矩阵,用于替换传统基于边的模块性矩阵(邻接矩阵),得到能反映网络结构的模块性参数,并将其应用于复杂网络的社团挖掘,根据得到的社团结构对节点的标签进行预测。在Flickr数据集上的实验结果表明,与Mod Max,Lable Diffusion和Edge Cluster方法相比,该方法对节点标签的预测性能较好,能更准确地反映网络的社团结构。

In community mining process of complex network,traditional modularity definition depends on the adjacent matrix of a network,and only takes the relationships between nodes into consideration,which cannot reflect the structure of the real network well. Aiming at this problem, this paper proposes a path-based modularity analysis method. It constructs the modularity matrix based on paths between any two nodes and their probability distributions, substitutes the edge-based modularity matrix,i, e. adjacent matrix, and gets the modularity value that reflects the network structure. Using the path-based modularity analysis method for mining the communities of a complex network,this paper predicts labels for unlabeled nodes according to the community structure. Experimental results on Flickr dataset show that, compared with ModMax, LableDiffusion and EdgeCluster methods, the proposed method has better prediction performance when predicting labels for unlabeled nodes, and it can reflect the community structure more accurately.