The Research And Design Of Overlapping Community Mining In Complex Networks

Complex networks, whose internal structure is complex and diverse is currently one of the hottest topics and has attracted a lot of researchers in different fields involving computer science, sociology and so on. There are many distinct properties in networks and community structure or cluster is a quite critical one, which performs as more links between nodes in the same cluster and few in different ones. Due to the network complexity, it is common to see overlap between different communities, which makes overlapping structure become one of the current focus taking the place of nonoverlapping community research. It proves to be more effective in detecting real-world network structure and revealing potential laws, which is also the main point of our paper.Overlapping cluster mining can be divided into two categories, including globalbased algorithms and local-based ones. Local detecting has an advantage over global one for it needs less calculation and does not need prior knowledge. However, it is becoming increasingly difficult with the increasing of network scale and the amount of data. It means improving the mining accuracy and reducing the time complexity are still the main objective of current research.In order to improve the quality of overlapping community mining, we design a local community mining algorithm based on the core nodes. It improves in both selecting initial nodes and mining communities. We select core nodes as mining center by calculating their comprehensive influence and design the adaptive function by considering both aggregation degree and density of the community so that the algorithm can be applied in more situations. The improved algorithm proves to be of higher mining quality comparing with other local algorithms.In order to improve the mining efficiency of overlapping community in complex networks, we design a mining algorithm based on core cliques. The analysis of node distribution in networks shows that overlapping nodes only account for a very small fraction of all nodes while traditional algorithms tends to calculate the fitness of all nodes repeatedly. To solve this problem, we dig the community starting with core cliques which is composed of core nodes and its closest neighbors, and adjust the distribution of cliques by measure the compactness degree of two nodes. After that, we first mine the main structure and identify the potential overlapping nodes of networks, and then only compute the fitness of these nodes to implement their multi-allocation. By this way, we sharply reduce amount of computation to increase the efficiency of the algorithm. It proves that the improved algorithm can detect the overlapping community in large-scale networks within acceptable time. Meanwhile, its mining quality is always higher or at least similar to that of other algorithms of the same kind.