Research And Implementation Of Local Community Detection Method In Complex Social Network

With the continuous development of the Internet,people’s daily life is full of various forms of social networks.The community structure usually reflects the cluster of closely connected nodes in the network.The community detection of social network plays a very important role in the tasks of studying the characteristics of the network structure and analyzing the potential laws of the network.However,the traditional global community detection methods are more and more difficult to apply in today’s social networks due to their large data scale and complex link relationships.At the same time,the rich node attribute information contained in social networks is often ignored in community detection.Therefore,how to realize the task of local community detection in complex social networks has become a research hotspot in the field of social network mining.In complex topological networks,existing local community detection methods generally have problems such as ignoring high-order topological information and sensitive seed nodes.In order to solve the above problems,combined with topological motif theory,a local community detection method LPRW(Label Propagation Random Walk)based on random walk and label propagation is designed and implemented.LPRW fully considers the topology information in the network,and implements a topological relationship weighting scheme for community detection.In the node sampling stage,the idea of label propagation and core nodes is introduced to improve the random walk method,which can help node sorting and community division more reasonably,and improve the accuracy of local community detection in topological networks.Most of the current research work on local community detection only focuses on network topology information,ignoring the rich node attribute information in real social networks.In order to effectively detect the community structure in social network,a local community detection method based on attribute network is designed and implemented.Firstly,attribute relationship is constructed according to the attribute similarity between nodes and their multi-order neighbors,and the LPRW method is used to sample the attribute and topological relationship to obtain candidate node sets.On this basis,a local community detection method LCDPSO(Local Community Detection by Particle Swarm Optimization)combined with multi-objective particle swarm optimization is proposed to iteratively screen the candidate node set,and obtain a community structure with compact topology and homogeneous attributes,which effectively improves the quality of local communities.Today’s complex social networks usually have the characteristics of large community scale and high attribute dimensions,which makes the execution efficiency of local community detection methods a problem that must be considered.In order to efficiently perform complex social network detection tasks and improve the detection efficiency of LCDPSO,a distributed computing-based local community detection method DLCDPSO(Distributed-LCDPSO)is designed.It implements a distributed design for particle swarms and performs parallel updates of particles with the help of the RDD data structure provided by Spark.At the same time,a particle design method for incremental computing is proposed,which realizes the fast calculation of the fitness function based on the previous iteration state and incremental changes of the particles.Finally,the proposed local community detection methods are integrated in today’s real social network analysis systems,effectively verifying their usability.