Research On Modularity Based Community Detection Algorithm In Large-scale Complex Network

Many of the complex systems existing in reality such as the Internet,biological networks,and social networks can be modeled as networks that contain nodes and their connections.The study of the topological relations of these networks will help us to find some imperceptible features,and then the network characteristics such as the similarity between nodes or the organization of the system can be more in-depth understanding.An in-depth understanding of the hidden relationships between nodes in the network,information flow analysis,and the organization and function of the system can be achieved by using community detection.The results of community detection can be used as the basis for further data mining.Therefore,it is of practical significance to study the community detection algorithm in depth.Although the predecessors have done a lot of work in the field of community detection algorithm,there are still some problems in the existing algorithms,such as the local community detection algorithm has lower accuracy,and the global overlapping community detection algorithm has higher time complexity,low accuracy and the lack of judgment conditions,and the current correlation matrix community detection algorithm does not consider a variety of relationships and so on.Considering the shortcomings and limitations of existing algorithms,several modified algorithms are proposed in this paper.The main work of this paper includes:In this paper,a two-stage local community detection algorithm based on maximum entropy random walk is proposed to solve the problem that the accuracy of the local community detection algorithm is not as good as that of the global algorithm.The modified algorithm firstly uses the maximum entropy random walk to detect the subgraph within a certain range of the source node,then obtains some nodes that have closer relationship with the source node as the seed nodes.The obvious advantages of the modified algorithm are that the maximum entropy random walk is used to improve the efficiency of seed selection,and the random factors are introduced in the fine search to avoid the local optimal solution.The simulation results show that the two-stage local community detection algorithm based on maximum entropy random walk improves the accuracy of community detection.In this paper,a parallel community detection algorithm based on local core extension is proposed,which is based on the issue that the traditional global overlapping community detection algorithms have high time-complexity and inflexible seed selection process and lack the post-processing steps.The process of the modified algorithm includes initial filtering of network,local core search,local core expansion and the post-processing.The modified algorithm avoids the subsequent useless processing through the initial filtering.Through the local core search,the initial seed selection is more uniform and the number of seeds automatically adapts to the network scale.The modified algorithm also establishes the interconnection between the global modularity and the local modularity.The simulation results show that the modified algorithm has better accuracy and quality than the existing algorithms.In this paper,a community detection algorithm based on multi-relational correlation matrix is proposed considering the existence of multiple behavioral correlations between individuals.The modified algorithm uses the stochastic matrix theory to filter the random noise of the correlation matrix.Then the modularity-based community detection algorithm is used to detect the community number of the single correlation matrix.Finally,the multi-relational correlation matrix is extracted,and the community detection is made with the matrix decomposition and the clustering algorithm.The multi-relationship is taken into account to improve the accuracy of the modified algorithm.