Research On Community Detection Algorithms In Social Networks

With the rapid development of information technology,the mining of social networks has been a very popular field of study in recent years.As a basic research of social network analysis,community detection has important implications for understanding the structural characteristics of the network and revealing the implicit patterns in the network.This paper investigates three aspects: improving the accuracy of community detection,improving the anti-noise performance of community detection,and identifying the overlapping community.In order to solve the problem of low accuracy when dealing the high-dimensional matrix with traditional clustering methods,we propose a community detection algorithm based on signaling process(SPC algorithm).In this algorithm,the influence matrix of the nodes in the network is obtained by using the idea of signal transmission,and the network's topological structure is transformed into the space vector information.Then,in order to improve the accuracy of community division,we design a deep sparse autoencoder to extract the feature of the high-dimensional influence matrix.The low-dimensional feature matrix can represent the community structure of the network better,and reduce the complexity of the clustering task.Finally,we use k-means method to cluster the low-dimensional feature matrix.The experimental results demonstrate that the SPC algorithm can improve the accuracy of community detection.Aiming at the problem that most of the community detection methods cannot be applied to the noise network,an algorithm of active learning for semi-supervised community detection is proposed in this paper(ASCD algorithm).This algorithm is a combination of semi-supervised community detection algorithm based on external optimization and a priori information select algorithm based on active learning.The semi-supervised community detection algorithm uses a priori information to guide the community detection which can effectively compensate the impact of noise.The active learning algorithm selects the nodes with the largest amount of information to enhancement the utility of priori information through the active learning method.The experimental results on real world networks show that the ASCD algorithm has a good anti-noise performance,and the experimental results on artificial networks verify the effectiveness of the active learningalgorithm.In order to effectively discover the overlapping community structure in social networks,we present an overlapping community detection algorithm using fuzzy c-means clustering(ODFC algorithm).The basic idea of ODFC algorithm is converting the network into a data structure suitable for cluster analysis,and then use the fuzzy c-means algorithm to detect the overlapping community.In the preprocessing of the network data,this algorithm takes into account of the power-law distribution of the node's degree,and proposes a new formula to calculate the node similarity.Then,the similarity matrix is mapped into two-dimensional space by multidimensional scaling,it's used as the input data of the clustering algorithm.At the same time,an initial clustering center selection scheme is proposed for the fuzzy c-means algorithm,which considers the node's importance and the edge's information center.The experimental results demonstrate that the ODFC algorithm can reveal the overlapping community structure in the network more accurately than other algorithms.