Community Detection Algorithm Base On Multi-Objective Particle Swarm Optimization

The network can be seen everywhere in people's lives.In the network,individuals often form communities through exchange of information.The discovery of the community is a basic research topic in the study of complex networks.It requires the community to have nodes with dense intraconnections and sparse interconnections.This can turn community discovery problems into multi-objective optimization problems,Multi-objective particle swarm optimization is an effective method to solve these problem.In dynamic networks,the topology of the network changes with time.According to this feature,this dissertation proposes a consensus-communities based multi-objective particle swarm optimization algorithm(CCPSO).This dissertation divides the consensus communities into two types:the consensus community within the population,which is the "knowledge" extracted from the high-quality solution at the previous time steps,reflecting the commonness of the network.The consensus communities among populations uses the "knowledge" of the previous time step as a guide.At this time step,it supports the degree of calculation and dynamically embeds the population,thus making the network develop toward the direction of the previous moment.The accuracy of the algorithm on the dynamic network is verified by artificial data sets and real-world data sets,and the rationality of the relevant parameters is verified by experiments.In static networks,the multi-objective particle swarm optimization algorithm uses an intuitive and simple adjacency matrix method,ignoring the hidden information in the network.In this dissertation,the network representation learning method is applied to the static network community discovery,and a multi-objective particle swarm algorithm(GAN-PSO)based on a Generative Adversarial Nets model is proposed.This dissertation simulates the real network structure by generating feature representations of the learning network nodes against the model.The resulting feature matrix is reused to improve the initialization process of the particle swarm algorithm and the calculation process of the objective function.Experiments in artificial datasets and real-world datasets have greatly improved experimental results,indicating that the algorithm is of practical significance.