摘要
随着社交网络的广泛使用,信息分享变得非常便捷,但这同时也加剧了虚假新闻的传播。现有多模态虚假新闻检测方法在融合文本与图像等多种信息源时,未能充分挖掘和利用新闻内容中不同形式的信息,导致语义信息使用不充分;此外,在融合不同层次结构的语义信息时未能充分考虑不同视角和层次间语义信息的关联性,影响了对新闻内容的深入理解。为此,该文提出了一种多视角特征融合的多模态虚假新闻检测方法,该方法从单模态语义、情感信息和多模态语义等多个视角对新闻内容进行挖掘,并设计了一个语义融合模块,有效整合来自不同视角和层次的语义信息,深入分析信息之间的语义关联性,以达到提升虚假新闻的识别准确性。在Weibo、GossipCop和PolitiFact等公开数据集上的实验结果表明,该文所提方法取得了优异的性能,比基准方法分别提升了1.4%、0.8%和4.6%。
Existing multimodal fake news detection methods fail to fully explore and utilize the diverse forms of information in news content when fusing multiple information sources,such as text and images,resulting in insufficient use of semantic information.Additionally,the correlation of semantic information between different perspectives and hierarchical structures is not sufficiently explored.To address these issues,this paper proposes a multi-perspective feature fusion method for multi-modal fake news detection.This method explores news content from multiple perspectives,including unimodal semantics,emotional information and multi-modal semantics.Furthermore,it introduces a semantic fusion module that effectively integrates semantic information from various perspectives and levels,capturing the semantic correlation between the information.Experimental results on public datasets such as Weibo,GossipCop,and PolitiFact demonstrate that the proposed method achieves improvements of 1.4%,0.8%,and 4.6%over the benchmark methods,respectively.
作者
肖聪
刘璟
王明文
王菲菲
邵佳兴
黄琪
罗文兵
XIAO Cong;LIU Jing;WANG Mingwen;WANG Feifei;SHAO Jiaxing;HUANG Qi;LUO Wenbing(School of Computer and Information Engineering,Jiangxi Normal University,Nanchang,Jiangxi 330022,China)
出处
《中文信息学报》
北大核心
2025年第4期126-137,共12页
Journal of Chinese Information Processing
基金
国家自然科学基金(62266023)
江西师范大学研究生创新基金(YC2023-s297)。
关键词
虚假新闻检测
多视角
语义融合
多模态
fake news detection
multi-view
semantic fusion
multimodal
