Research On Algorithm Of Community Structure Analysis Based On Complex Network

Detecting community structure of complex network have important clinical implications for understanding the function of complex networks, analyzing the relationship between the nodes, understanding the features of topological structures forecasting the future behaviors of complex networks. Although there are a lot of methods have been proposed at present, but how to further improve the clustering accuracy, especially how to discover reasonable community structure without prior knowledge such as the number of communities, and how to detect the community structure in large-scale complex networks quickly are still challenging research problems.For the complex network community structure discovery, swarm intelligence algorithm has become one of the hottest research topics. Ant colony clustering algorithm is a swarm intelligence optimization algorithm. It is a simulation of the ant behavior such as cleaning nest or looking for food. And it possesses many characteristic like dynamic, self-organization, high efficiency, high clustering quality etc. Based on research experience of the domestic and foreign scholars on the community structure in complex networks and ant colony clustering algorithm, this paper mainly do the following work:1) For the small and medium-sized network, considering the node attribute information, using the model based on node similarity and diffusion model to improve ant colony clustering algorithm. In the algorithm we use ant directly as a node in the network. In the algorithm, each ant represents a node of a complex directly network. At the beginning, ants are randomly placed at different locations in a virtual grid. During clustering, ant decide to move or stay by calculating the similarity to the surrounding environment and the degree of adaptation. Each Iterative of ant colony moving will form a clustering result. At the same time, we use the pheromone diffusion model to the algorithm, and it updates according to the clustering solution. At the same time, according to the adaptability of ants and the environment, ants get their distance to move. The above process repeat many times until all of the ants having the right positions. The community structure would show in the grid.2) For large networks, the speed of clustering method is slower in large scale. Sampling is carried out using the key nodes in the network. It reduces the scale of the problem. And we lays the thought of label propagation to sign the sampled nodes and cluster sampling nodes using the above work. Finally, the remaining nodes is assigned to the community according to the similarity of nodes and the community. Algorithm uses of modularity function as the index to modify the clustering result.