Research On Community Mining Algorithms In Complex Networks Based On Clustering And SOM

With the rapid development of computer technology,it's a necessary way to meet the needs of material and cultural by collecting and resolving large-scale and wide ranged actual data from network.Network science had also played an increasingly important role.It has been close to people's lives,such as social networks,biological networks,information networks,transportation networks,and so on,it has been intertwined with each other.It is a main point of internet research that reveal the common issue in the network and the universal methods of solving these problems,and these networks can be summarized in the scope of complex networks.It is the excavation of the hidden community structure that plays a vital role in the prevention of viral transmission,the control of public opinion,and the prediction of the function of the unknown biology.The main contributions of the dissertation on mining the community structure in complex networks are summarized as follows:(1)This dissertation summarizes the current status,related definitions,properties and models of complex networks,and analyzes the hierarchical division of community structure.In addition,describes the significance of studying the structure of community,summarizes the advantages and disadvantages of typical community structure division algorithm,narrates the thought of clustering algorithm and self-organizing competitive neural network(SOM for short)related knowledge to excavate the community structure.(2)A community discovery Algorithm Based on Shortest Path Feature(SPCDA)is proposed.Based on the characteristics of the shortest path,the mediation coefficients of each node are calculated by the number of features to obtain the center of community,and the similarity between the nodes is calculated according to the characteristics of the length.A threshold is agreed as a rule for partitioning that is ultimately determined by the average similarity of all nodes.And then constitute a model similar to the cluster,finally,in accordance with the partition rules that each node(except the center nodes of the community)were compared with the threshold,take the node of the exceed threshold classification clustering,according to the process of constant iteration,until the completion of the division applied to the classical complex network experiment simulation platform,and compared with the typical GN algorithm and LPA algorithm.The results show that the SPCDA algorithm can quickly and accurately excavate the hidden community structure.(3)A multi-feature community algorithm based on self-organizing competitive neural network is proposed(SOMCDA).Considering the topology of the network and the node feature attribute,the clustering idea is combined with the SOM.The algorithm proposed in this chapter is based on the influence of nodes,and the cohesive coefficients of each node are calculated by combining the degree of nodes and the number of edges between adjacent nodes.The feature nodes are extracted from nodes with large cohesive coefficient values.These representative feature nodes are used as sample nodes.Then,the multi-feature attribute information of the sample node is trained by SOM,and the non-sample node is provided to the trained SOM.According to the characteristics of the SOM structure storage model,the competition network will make the identification,so as to achieve the purpose of community division.Finally,according to the number of competing neurons in each simulation,the modularity function is used to determine the best community structure.