Design And Implementation Of Large-scale Community Detection Parallel Algorithm On Complex Networks

With the continuous development of computer and Internet technologies,networks have become an indispensable part of people's daily lives.Most of the data in people's real life can be represented by networks or graphs,such as communication networks,transport networks,social networks,biological networks,and so on.Community structure is an important structural feature of complex networks.Exploring these community structures is conducive to a deeper understanding of the functions and characteristics of complex networks.In addition,complex networks often show features at the community structure level,which are not available at the individual nodes or at the entire network level.In order to explore complex networks deeply and capture more meaningful potential information,community detection techniques are necessary.However,as the number of nodes continues to increase and the network structure continues to be complex,traditional community detection algorithms are no longer suitable for large-scale complex networks.Distributed parallel processing technology provides an effective solution for processing large-scale complex networks.Therefore,the research for parallel community detection algorithms on complex networks has great theoretical significance and practical value for the in-depth understanding of network structure and features.Based on the above reasons,this paper proposes a parallel community detection algorithm for large-scale complex networks,Parallel LPA-Attractor algorithm(abbreviated as PLA).The process of designing this algorithm involves graph representation,partition,storage,calculation and result fusion.The most important one is the community detection algorithm of a single computing unit.After comparing and balancing the existing community detection algorithms,the Attractor algorithm is used as the core computing algorithm.However,this algorithm will detect a large number of single-node communities.The LPA-Attractor algorithm(abbreviated as LA)proposed in this paper improves the problem by using the label propagation algorithm LPA.As the LPA algorithm can only find non-overlapping communities,this paper proposes an improved algorithm,Adjustable Multi-Label Propagation Algorithm(AMLPA),so that the node carries multiple labels to detect overlapping communities.In other aspects of algorithm parallelization,the existing graph representation method is no longer suitable for parallel and large-scale networks.This paper presents a graph representation of the coupling table.Coupling tables allow data to be used more efficiently and speed up calculations.The PLA algorithm uses the METIS graph partition algorithm to partition the input graph data.In terms of data storage,this paper proposes an improved distributed storage method that uses the deep replication of nodes to solve the problem of large communication cost of existing distributed storage methods.In addition,an effective fusion scheme is proposed for parallel processing results.The results of relevant experiments show that the parallel community detection algorithm proposed in this paper effectively reduces the computation time and has the advantages of lower time complexity and higher accuracy.Among them,the core computing algorithm LA greatly reduces the number of single-node community.Therefore,the PLA algorithm provides an effective solution to solve the parallel community detection problem for large-scale complex networks.