Study On Community Detection Using Deep Network Representation

With the fast development of computer network and increment of active users,the study of complex networks has received an enormous amount of attention in recent years.Community detection in complex network can help to analyze and classify the users in large-scale network platform,accordingly each researcher gives definition to community and propose different methods to solve.Spectral methods are kernel mathematic methods,which indeed is a lower rank representation of original data.In many conditions,under the relaxation of constraint,the optimization of object function can transform to a spectral method,reducing the complexity of the optimization.Deep learning is a hot topic nowadays and it is characterized by extracting the input data features.Take the similarity of deep learning and spectral methods into consideration,we propose a replacement of spectral methods with deep representation for community detection.Based on the modularity model and normalized cut model,we apply the autoencoder to represent the features of modularity and Markov matrix,the results of which are clustered by K-means.The experimental results show that deep representation can the place of spectral methods,indicating a great value in community detection.