Community Detection Based On Convolutional Neural Network

Many systems in the real world can be modeled as networks(graphs)G(V,E).The individual entities in these networks are represented by a set of nodes V(vertices,|V| = n)and relationships between these entities by a set of links E(edges,|E| = m).For example,in the metabolic network of Biology,nodes are proteins and links are the chemical interactions between them.In social networks,nodes represent individuals and links correspond to the relationships between them.The complex network has a key property,which is the community structure,and the most important characteristics of community structure is that the nodes within the same community are tightly connected,but the nodes between different communities are relatively sparse.Therefore,detecting the community structure of networks can enable us to better understand the topological properties and organizational structure of complex systems,which is of great practical significance for us to understand the real world.To date,many community detection algorithms for complex networks have been proposed,including spectral graph partitioning,hierarchical clustering algorithms,modularity maximization algorithms,label propagation algorithms and random walk algorithms.However,because most of these algorithms are based on evolutionary algorithms for numerical optimization,they exhibit slow convergence and poor local search ability,making them less ideal for large-scale network community detection.In this paper,based on convolutional neural networks,we propose a community detection algorithm using the aggregation of complex networks and the characteristics of deep learning systems which can effectively handle big data classification.This algorithm can effectively remove the edges between the different communities in the network,thus mining the real community structure of the network,especially for large-scale network systems.The contributions of this paper are as follows.(1)Putting forward a modeling method(Edge to Image,E2I)to convert the edge structure information in the network into color images,based on the aggregation of complex networks.(2)Constructing the convolutional neural network model(Com Net)to indirectly classify edge types(within or between communities)in the network by virtue of the characteristics of its classification images,and to remove the edges between communities to get the preliminary communities on the basis of edges classification.(3)Using local modularity R to merge the preliminary community structure and optimize the final community division results.The algorithm proposed in this paper is essentially a splitting algorithm using graph neural networks,which uses the trained CNN model to classify the edges in the network and split the network.This process effectively utilizes the local information of the network,and can split the network once in the whole network scope,reducing the iterative process and improving the efficiency of the algorithm.Experiments on real networks and computer-generated networks show the feasibility and effectiveness of the proposed algorithm.