基于多模态动态图学习的假新闻检测研究综述

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）12-002-3534-09

doi：10.19734/j. issn.1001-3695.2025.04.0136

Review of research on fake news detection based on multimodal dynamic graph learning

Yang Yulat，Guo Ruilb，2，Gao Minna³，Wang Yifanla，Lu Yaolin 1b，4 （20 （1.a.Schoolfn，GdateentgdeUstfi';tf InformationCouicationCneseeople'sedPiceceocommisioedOficerShool，HagzouO，hiocet ForceUniersityfgering，i’8hina;4obileehentfnedoliceCos，z0a

Abstract：The proliferationof social media accelerates the spreadoffakenews atanunprecedented speedand scale，posing a significanttreattosocialstabilityandpublictrust.Traditionaldetectionmethods，whichreliedontextualcontentalone， strugled to counter modern AI-generatedfake news characterized byrich media，diverse formats，andrapid dissemination. This papersystematicallsurveyedthe technicalpipelineforfakenewsdetectionandanalyzedrepresentativemethodsandresearch progressacrossfourkeystages：data perception，feature analysis，path tracing，andactive blocking.It examinedcore challenges，，including multimodaldataprocesinganddynamic sourcetracing，whilesummarizingthemeritsanddrawbacksof variousapproaches.Thestudy highlightedmultimodaldynamicgraph learningasakeyenabling technology，leveragingits strengths incorrelatingmultimodalinformationandcapturingthe temporal evolutionofpropagationnetworks.Specifically， multimodal techniquesofer profoundapplications fordataperceptionandfeature analysis，whiledynamic graphlearning accuratelycaptures the propagationcharacteristicsoffakenews，enabling moreeffective path tracingandultimatelyachieving more precise and efficient active blocking.

Key Words：fake news detection；multimodal fusion；dynamic graph learning； lightweight design

0 引言

“假新闻"（fakenews）的概念在2016年美国大选期间被广泛关注，Allcott等人[1]将其定义为“故意并可验证为虚假的、可能误导读者的新闻文章”，带有一定的政治色彩且具有煽动性[2]。（剩余27945字）