Research On Community Detection Algorithm In Complex Networks

The research of complex networks is very helpful for people to understand the behavior of individuals in complex systems,and reveal the implicit phenomena that are seldom have been paid attention to it.In recent years,community structure has received extensive attestation as an important part of the study of complex networks.This paper is combined with the current studies and existing problems,and the main innovative work and research results are summarized as follows:1)To design community detection algorithm based on the node correlation strength.According to the number of the neighbors of the node and the weights of the edges between the neighbors,the probability distribution function is used to determine the expectations of the nodes and the pairs of nodes to be selected,and the correlation of the corresponding pairs of nodes is calculated.The purpose of community division is achieved by dividing the two nodes with greater correlation into the same community.Experimental results show that compared with the traditional algorithm,the proposed algorithm can ensure that the number of communities is in line with the actual number on the basis of a higher modularity.2)A conversion algorithm is proposed,which used to convert a directed network into an undirected network.The interaction between individuals in the real world has the meaning of goodwill or malice,meanwhile,the interaction is directed.Hence,we propose an affinity index to measure the popularity of individuals in the network.And then update the weights of edges iteratively using the affinity index to make the network reach or approach the dynamic signed consistency state.Finally,the directed network that under the dynamic signed consistency state will be converting to an undirected network.3)The improved signed modularity index and the community detection algorithm for undirected signed networks are proposed.Although some scholars have proposed an improved modularity function suitable for signed network,it calculates the modularity by weighted summation,while the sign properties of signed network are not taken into account.Therefore,we propose an improved signed modularity function which can represent the sign characteristics of signed networks,and a community detection algorithm based on reconstructing networks,the scale of the network will be reduced during the execution of the algorithm.The experimental results show that the performance and efficiency of the new algorithm are not worse than the existing algorithm,and the results given by the new algorithm are consistent with the actual situation.4)To study the signed local modularity index,and propose a community detection algorithm based on local information and dynamic expansion.It is difficult or even impossible to obtain the global information when the network is dynamic or large-scale.Therefore,it is necessary to use local information to detect the community.First,we propose a signed local modularity index to measure the quality of local communities in order to achieve the global optimization by making the quality of each local community to reach the optimal.As shown in experimental results,the proposed algorithm gives the same results as the existing algorithm,which indicators that it is feasible to mine the community structure by using local information and has more advantages in recommendation field.