Research On The Abnormal Behaviors Of The Network Based On The Complex Network Theory

Studies about social network analysis and complex network theory existed haveshowed that, based on the network theory’s capability to describe both the power-lawdistribution and the small-world effect in the social networks, the societal computing canmeasure the complex behavior in the society. The universal nature of the random network,which obtained by probability theory, ensured the correctness of the method that quantifiedthose abnormal behaviors in the network with stochastic number.A scale-free network is a network whose degree distribution follows a power law. Themost notable characteristic in a scale-free network is the relative commonness of verticeswith a degree that greatly exceeds the average. And the distribution is continuous, so athreshold does not exist to distinguish the high-degree nodes and low-degree nodes. Firstlythe statistical method is proposed to distinguish two types of nodes with different activedegrees through the character of the power-law distribution，which can be deployed asspam filter.Considering the small-world effect, the average distance between any two nodeschanged lightly while the amount of nodes varied significantly. The tree model does notmeet the power law distribution, where this character also could be observed. This featureimplies that the new way to construct a graph is feasible. In this way, the shortest averagepath is determined fisrtly, and then the nodes are added according to certain rules.Secondly the communities widely discovered in network are important to the socialanalysis. With the help of community searched algorithm, a certain presumption aboutwhether the network is community-structured is given. It calculates the proportion of thenumber of edges within the community and the number of edges between the communities.And the self-similarity coefficient of the certain community is given by the spectraldecomposition of the matrix which corresponds to the community. According to comparingthe two self-similaritiy coefficients of any two communities, the presumption of whethercommunities are similar is done. The social network is gradually formed with the constant movement of nodes andedges. Then the evolution model of the community has been proposed. This simulationprocess of two communities at different scale level merged into one was completed. And aconjecture about all that the process of two communities are merged into one needed isquite a few edges between those communities is proposed along with this model.The characters of different active nodes were validated in the spam filtering procedurefrom the original electrical mail records at last. According to a set of analytical programs, itis not surprise to find that those spam nodes also belong to the botnets. Then thecomprehensive analysis platform which detects the botnet of distributed communicationmechanism is desired.