基于模块度增强与双监督嵌入的社区发现算法

打开文本图片集

中图分类号：TP18 文献标识码：A 文章编号：1006-8228（2025）09-16-07

Abstract：Incomplexnetworkanalysis，communitystructureisanimportantcharacterizationofnetworkfunctionsandorganization Existingcommunitydetectionmethodssufferfromisuessuchasneglectingigh-orderstructures，dificultyineplyfusing atributeandstructuralinformation，andsuboptimalsolutionscausedbytheseparationofembeddinglearningandclusterigtasks. Toaddresstheseproblems，acommunitydetectionalgorithmbasedonmodularityenhancementanddual-supervisedembeding （MEDSE）isproposed.Thealgorithmextractsnodeatributeinformationthroughaselfatentionencoder，reconstructsthgraph convolutionalnetworkusinganon-negativemodularitymatrixtoenhancetheabiltytoextracthigordertopologyfusesatribute andtopologicalinformation，andaleviatestheover-smothingproblemoftraditionalGCN.Adual-supervisedmechanismis employedtoguidetheunsupervisedclusteringofthemodel，makingnetworkembeddingorientedtowardsthecommunitydetection task.Comparativeexperimentswith14benchmarkalgorithmsshowthattheMEDSEalgorithmcanefectivelyperformcommunity detection on different datasets with higher accuracy.

Keyords：NetworkEmbedding;CommunityDetection；Self-AtentionEncoder；GaphConvolutionalNetwork；HigherOrderStructure

0引言

在复杂网络分析中，社区结构是网络功能单元的重要表征，对理解网络的结构和功能具有重要意义。（剩余10007字）