Detection And Visualization Of Hierarchical Overlapping Communities In Complex Networks

For discovering the structure features of complex networks, many researchers try to detect the communities in the networks. Finding out the realistic communities is significant in analyzing and using the networks. This paper proposes an algorithm for detecting hierarchical overlapping communities, and visualizing the detecting results with a hierarchical genetic algorithm.Firstly, the detecting algorithm is used to optimize the community structure locally in order to find out the overlapping nodes from the partition which is the result of maximizing the modularity of a complex network. Then a graph which indicates the relationships between communities is built upon the cover of a network. By building the community graph and detecting the community iteratively, the hierarchical overlapping communities of the network which represents the features of a real system is detected. All of the community structures, overlapping structures and hierarchical structures are beneficial for understanding and using the complex networks (e.g., to control the networks and discover the new functions).According to the idea of dividing and conquering, a top-down hierarchical genetic algorithm is proposed to visualize the results of the detecting algorithm. By layouting each community and replacing the community nodes with sub-layouts, the approximate best layout of the network is achieved. To gain global optimization through accelerating local optimization process, the rotating, merging and overturning processes are introduced. The visualization which takes the advantage of the structure information of a network is very useful for observing and analyzing the network structure clearly and directly. The computational efficiency is satisfying especially in a parallel setting.The experimental results prove that, the hierarchical overlapping community detecting algorithm is feasible and promising. And given some rules, the hierarchical genetic algorithm can achieve a high efficiency and accurately visualize the network structure.