基于传播图主题解耦的微博谣言集成式检测方法

打开文本图片集

关键词：微博谣言；传播图；主题解耦；图表示学习；集成式检测

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2026）01-005-0043-10

doi：10.19734/j.issn.1001-3695.2025.06.0206

Integrated detection method for microblog rumor based on topic decoupling of propagation graph

Xu Jianminlat，Wang Qila，Zhang Xiongtao 1b，2 （1.a.ScholofyburitydCouterhoolofageent，beUiesitodingHbeioina;.off mationManagement，NanjingUniversity，Nanjing21oooo，China）

Abstract：Recent researchon microblog rumor detectionuses structured information to learn the dissemination process of microblogsandenhance theirrepresentation.However，existingmethodsignorethediferences inpropagationacrossdiferent topics，which leads toinaccurate modelingandafectsdetectionperformance.To solvethis problem，the paperproposedanintegrated rumordetectionmethodbasedontopicdecouplingof thepropagationgraph.Themethod firstlyconstructedamicroblog propagationgraphbasedonrepostingrelationshipsand initializedthegraphusing textcontentanduser information.Then， itappliedavariationalautoencodertopicmodeltodecouplethetopicsinthepropagationgraph.Italsoappliedaself-attention graph poling mechanism tolearntopic-levelrepresentationsof the graph.Finally，itdesignedahierarchical ensemblemechanismtodetectrumorsfrombothlocaland globalperspectives.Experimentsonpublicmicroblogdatasetsverifytheeffectivene andrationalityof the proposed method.Compared withbaseline methodssuchas BiGCN，MGLAN，andFGCN，the proposed method achieves significant improvementsinaccuracy，recall，and F1 score.The experimental results demonstrate that the method performs effectively in rumor detection tasks.

Keywords：microblog rumor；propagation graph；topicdecoupling；graph representation leamning；integrated detection

0 引言

随着在线社交的发展，微博平台已成为用户发布、传播信息的重要渠道，其在促进信息交流的同时，也导致大量网络谣言滋生[1]。（剩余25419字）