Design Of Community Detection Methods Based On Nodes' History Behavior In Social Networks

Community structure with high quality is very important to study the structure and function of the network. It is also helpful to find out the hidden rules and predict the possible behavior. It is of great theoretical significance and wide application prospect. This is the basis and key of the network analysis. It is also the focus of many researchers.There are many research results on community detection at present.However, there are still some limitations and shortcomings in the practical application of current research findings, such as the input of prior information (the number of communities, the number of nodes in each community), the instability of community division, and the low quality of community. Mainly because most of the methods do not take full use of the behavior characteristics of nodes into community detection,or only pay attention to algorithm operation speed while neglecting the quality of community structure. Therefore, the quality of the community structure is lower. Therefore, we should consider fully the characteristics of the nodes' historical behavior and develop the efficient detection algorithms.Based on the structural characteristics of the network, we divide the social network into static networks and dynamic networks. We will study community detection algorithms for both.In the static network, some algorithms depend on the input of prior information. Some algorithms can only detect non-overlapping community. The result of community structure detected by some algorithms is not stable or the quality of community structure is poor. To solve such problems, we consider the relationship strength between nodes and the modularity of the community, and propose an overlapping community detection algorithm via maxing modularity of communities.Firstly, the algorithm analyzes the movement of nodes based on the node's historical behavior and compute the relationship strength between nodes, thus we get an undirected weighted network. Secondly, it sorts edges of the obtained network diagram by size and selects nodes with highest relationship strength as the initial community. The higher the strength of nodes, the more stable the relationship between the two nodes is. Nodes with higher relationship strength are much easier to form a community and the location of the nodes in the community is closer to the center. Finally, the initial community expands based on the optimization of modularity. New nodes are involved into current community until it reach the termination condition.In the dynamic network, considering that in a short time small-scale network changes will not make significant changes in the network structure, only a few related nodes and their adjacent nodes may change the membership of the community. The network structure is generally stable. We propose a community detection algorithm based on network increment in the dynamic network. Firstly, it inducts the increment of the network caused by nodes' behavior into six atomic operation. They are adding and deleting isolated nodes, adding an inner edge included in community, deleting an inner edge included in community, adding an outer edge out of the community, deleting an outer edge out of the community and changing weight of the edge. Secondly, it gives the detailed processing methods to the six atomic operations respectively and analyzes the influence of the nodes on the communities and adjacent nodes. Finally, it analyze the result set of all the above operations synthetically. It decreases the complexity of the algorithm and achieves a real-time network community efficiently.Through simulation experiments compared with classical similar algorithms, we prove that the two community detection algorithms have high evaluation accuracy for different types of networks.