Improvement Of Louvain Community Detection Algorithm And Its Application

Many systems can be represented by complex networks in the real world.Research on the structural characteristics and embedded mechanisms of complex networks has become an important way for people to correctly understand complex networks and promote the development and application of complex networks.With the continuous research of complex networks,people have discovered that community structure is an important feature in complex networks,and it helps to understand the structure and function of the entire network.With the increasing concern about the network community structure,many algorithms for dividing the network community structure emerge one after another,and its practicability and application have received more attention.This paper studies the improvement and application of community discovery algorithms in complex networks.The main work includes:(1)In order to eliminate the phenomenon of modularity oscillation and community division oscillation in Louvain algorithm,an improved calculation method of modularity gain is given,and we propose a community detection algorithm,called PRULOU.The algorithm includes four main processes: modularity gain(new(35)Q)calculation,node movement,pruning,and network aggregation.By judging whether the current node is consistent with the community label of the neighborhood,and then pruning the nodes in the network,this significantly increases the proportion of effective nodes in the node sequence.On 8 real networks,the proposed algorithm is compared experimentally with 7 classic algorithms such as Louvain,Fast Algorithm,and Statistical Monitoring and other algorithms.The results show that the PRULOU shows superiority in terms of modularity and time.(2)Aiming at the problem that the PRULOU algorithm can only bedivided based on the limited amount between nodes and communities,a pruning algorithm based on modularity and similarity measures is proposed,called PRUJMS.This algorithm gives a measure index JSM for calculating the similarity between nodes,and measures the similarity in combination with the gain of the modularity,so that the nodes can move more efficiently during the division process.Through experimental comparison and analysis with some classic algorithms on 8 real networks,the results show that PRUJMS performs superiorly and performs well in terms of modularity and time.(3)A comparative analysis system for community discovery algorithms,the Network Repository platform,is designed and implemented.It includes three functional modules: network data integration,community discovery algorithm integration,and evaluation index integration.Currently,the platform integrates 9 data sets such as Karate,Dolphin,Football,and Yeast Network.This network data is preprocessed to obtain statistical information such as network scale,number of nodes,number of edges,aggregation coefficient and other indicators.The community discovery algorithm module integrates and visualizes 5 classic community discovery algorithms(Grouping Algorithm,Clustering Algorithm,Fast Algorithms,Local Resolution-limit-free Potts Algorithm,and Statistical Mechanics),and gives each in the form of a histogram and linechart.Comparison results of the algorithm in indicators such as NMI,ARI,AC.