0352022年第6期/第39卷/总第210期电子商务水军检测的新方法:自适应邻域精准化采样的多关系图神经网络ANewMethodforDetectingE-CommerceSpammer:Multi-relationshipGraphNeuralNetworkUsingPreciseSamplingandAdaptiveNeighborhood徐瑞卿张志旺孙宏亮XURuiqingZHANGZhiwangSUNHongliang(南京财经大学信息工程学院,南京,210046)摘要:[目的/意义]旨在从图神经网络的视角提出一种新的水军检测算法,为保障电子商务环境健康、商家信誉、市场公平提供支持。[研究设计/方法]结合多关系图神经网络,引入新型采样策略,设计出一种基于精准化采样和自适应邻域的多关系神经网络的电子商务反欺诈算法,并将这种新算法应用于真实世界Yelp和Amazon的数据集上进行效果检验。[结论/发现]与过去的反欺诈方法对比发现:这一新方法在缓解类别不平衡带来的影响时有显著的效果。[创新/价值]该方法提供了一种新的抽样策略,为有效解决欺诈检测研究中面临的海量用户中仅有少量欺诈用户导致的类别不平衡问题,提供了一种新的思路。关键词:欺诈检测;类别不平衡;精准化采样;自适应邻域;多关系图中图分类号:G203DOI:10.13366/j.dik.2022.06.035引用本文:徐瑞卿,张志旺,孙宏亮.电子商务水军检测的新方法:自适应邻域精准化采样的多关系图神经网络[J].图书情报知识,2022,39(6):35-44.(XuRuiqing,ZhangZhiwang,SunHongliang.ANewMethodforDetectingE-CommerceSpammer:Multi-relationshipGraphNeuralNetworkUsingPreciseSamplingandAdaptiveNeighborhood[J].Documentation,Information&Knowledge,2022,39(6):35-44.)Abstract:[Purpose/Significance]Toensurethehealthofthee-commerceenvironment,businessreputationandmarketfairness,thispaperaimstoproposeanewspammerdetectionalgorithmfromtheperspectiveofgraphneuralnetwork.[Design/Methodology]Combinedwiththemultiplerelationalneuralnetwork,weintroduceanewsamplingstrategy,anddesignanewmethodofmultiplerelationshipgraphneuralnetworkbasedonprecisesamplingandadaptiveneighborhood.ThenthisnewapproachisusedontherealworldYelpandAmazondatasetstotesttheeffect.[Findings/Conclusion]Comparedwiththeexistingfrauddetectionmethods,thisnewalgorithm,hasasignificanteffectinmitigatingtheimpactofcategoryimbalance.[Originality/Value]Thismethodintroducesanewsamplingstrategytosol...