Community Detection Algorithm Based On Deep Learning

The development and popularization of information technology have brought people into a new internet era.More and more systems in life can be abstracted into complex networks.The research of complex networks has become a hot spot for information field.Community structure is one of the most important characteristic of complex networks.Discovering the community structure of complex networks can help us to explore the internal structures and attributes of networks,so we can discover the hidden rules of them.Therefore,it is of great significance to research the community structure of complex networks.This paper first introduces the concept and research status of complex networks,and summarizes the different types of community discovery algorithms.Among them,K-means clustering algorithm has been widely used in community discovery because the principle of it is simple,so we can easy to implement and it is suitable for various types of data.However,the algorithm is not sensitive to the high dimensionality and sparsity of complex networks,and the results of the community are not precision.Therefore,based on the classical clustering algorithm,this paper uses deep learning model to make full use of the prior information of the network.We propose the improved community discovery algorithms for linked complex networks and converged complex networks respectively.They mainly include the following contents:(1)We propose a deep learning-based community discovery algorithm for the link network named deep-CLCD.After that we propose a new node similarity calculation method,it can construct feature matrix based on the distance relationship between nodes and neighbor nodes.Facing the complexity of the network architecture,this paper uses the nonlinear mapping ability of deep learning and extract the effective characteristics to low-dimensional space;then we use clustering algorithm to return results.(2)An active semi-supervised clustering community discovery algorithm ESCD is proposed for the link network.In this paper,the traditional K-means algorithm is calculated step by step;the distance iteration result of each step is regarded as the rough clustering result,and the node membership degree is calculated according to the coarse clustering result;the network reconstruction is completed by actively adding the prior information.To make the network structure clearer,we use the clustering algorithm to return the result in the end.(3)A community discovery framework based on deep learning,CGLCD,is proposed for the content and link fusion network.The text content of the nodes in the network can be used to guide the network topology reconstruction,and the integrity of the entire network topology can be improved by constructing a node group with higher similarity;We propose a text and link fusion method,and the minimum and maximum of the threshold are set to guide the reconstruction of the topology by using the textual information in a pairwise constraint;the community discovery results are obtained by using the CLCD algorithm.In this paper,all algorithms are experimentally verified in a wide range of real data sets and artificial datasets.The experimental results show that compared with the existing community discovery algorithm,the deep-CLCD algorithm can discover the community more accurately;the ECSD algorithm can achieve great performance with fewer labels;the CGLCD algorithm performs well when dealing with text-fused link data sets.