Research On Community Detection In Networks Based On Clustering

Many complex systems in real world can be abstracted as complex networks,and the objects in systems can be represented by nodes in networks,the connection(edge)between nodes represents the relationship of the objects.The feature of the nodes represents the interest tendency of different objects.Community detection technology aims to discover the structure of complex networks,through which the nodes with high correlation are merged into clusters.We can analyze the significance of this classification in different fields.Researches show that community detection methods mainly employ clustering algorithm.In this paper,we mainly focus on non-overlapping communities and overlapping communities with feature.It is difficult to determine the number of communities only through the networks' original nodes information.The randomness of selecting similarity measure results in uncertain results of community patition.Also,greedy strategies in some optimization algorithms often lead to local optimal solution.Upon framework of information theory,we proposed partitional Information Bottleneck clustering based community Detection(pIBD)to solve problems above.Firstlty,we describe the mutual information loss curves,from which the optimal number of communities is predicted.The mutual information between clusters is used as the clustering similarity,on the foundation of which we design the objective function of partitional clustering process.Experiment results show that the pIBD algorithm can predict the number of communities by using the original nodes information,it also can improve the quality of community identification.The traditional methods of overlapping community detection mainly incorporate fuzzy clustering and co-clustering methods,generally acting on two-dimensional data networks.There is few clustering method for three-dimensional data networks with feature.Based on the methods of fuzzy clustering,co-clustering and three-dimensional clustering,we proposed an innovative fuzzy tri-clustering community detection algorithm.This algorithm analyzes the probability of both nodes relationship and their feature,and performs clustering on all three dimensions simultaneously.Experiment results show that the FTC algorithm has better clustering quality than those two-dimensional fuzzy clustering methods.The algorithm identifies the reasonable patition of overlapping community.The FTC algorithm also provides an effective approach for multidimensional data clustering.