642023ＲadioEngineeringVol.53No.1doi:10．3969/j．issn．1003－3106．2023．01．009引用格式:陈娜．基于图神经网络的复杂网络关键节点检测算法［J］．无线电工程，2023，53(1):64－72．［CHENNa．AlgorithmforDetectionofCrucialNodesinComplexNetworksBasedonGraphNeuralNetwork［J］．ＲadioEngineering，2023，53(1):64－72．］基于图神经网络的复杂网络关键节点检测算法陈娜(山西工程科技职业大学计算机工程学院，山西晋中030619)摘要:针对复杂网络关键节点检测算法准确性低及可靠性不足的问题，结合图神经网络(GraphNeuralNetwork，GNN)模型提出了一种新的复杂网络关键节点检测算法。将复杂网络建模为图模型，通过注意力机制学习每个邻居节点的权重;利用GNN强大的图学习和推理能力，评估网络中节点与连接的关键性评分;采用强化学习(ＲeinforcementLearning，ＲL)搜索GNN的超参数，从而提高关键节点检测算法的可扩展性及可靠性。仿真实验结果表明，由该算法检测的关键节点具有较高的准确性，并且具有较快的运算速度。关键词:复杂网络;关键节点检测;网络关键节点;社区检测;深度学习;深度神经网络中图分类号:TP391文献标志码:A开放科学(资源服务)标识码(OSID):文章编号:1003－3106(2023)01－0064－09AlgorithmforDetectionofCrucialNodesinComplexNetworksBasedonGraphNeuralNetworkCHENNa(CollegeofComputerEngineering，ShanxiVocationalUniversityofEngineeringScienceandTechnology，Jinzhong030619，China)Abstract:Todealwiththeproblemsoflowaccuracyandreliabilityoftraditionalcrucialnodesdetectionalgorithmsforcomplexnetworks，anewcrucialnodesdetectionalgorithmforcomplexnetworksisproposedbasedongraphneuralnetworksmodel．Firstly，thecomplexnetworksaremodeledasgraphmodel，andtheattentionmechanismisutilizedtolearntheweightofeachneighbornode;then，thepowerfulgraphlearningandinferentialabilitiesofthegraphneuralnetworkareused，asaresult，thecrucialityscoresofnodesandlinksinthenetworksareevaluated;finally，thereinforcementlearningisadoptedtosearchthesuperparametersofthegraphneuralnetworksinordertoimprovethescalabilityandreliabilityofthecrucialnodesdetectionalgorithm．Simulationexperimentalresultsshowthatthecrucialnodesdetectedbytheproposedalgorithmaremoreaccurate，atthesamet...
