基于图同构网络的多模态虚假新闻检测

Multimodal Fake News Detection Utilizing Graph Isomorphism Networks

摘要: 多模态虚假新闻检测是指通过整合来自多个模态的信息，识别和分类虚假新闻的一种技术。本文基于当前虚假新闻通常涵盖文本、图像以及复杂的社会关系等多重信息，且现有模型处理社交上下文信息结构信息捕捉严重不足，本文提出基于图同构网络的多模态虚假新闻检测框架(Graph Isomorphism Networks Fake News Detection, GIN_FND)，通过提取新闻中文本、图像和社交上下文信息等多模态信息，基于图同构网络提取社交信息的方式有效捕捉社交图节点的局部依赖和全局关联，有效提升模型检测准确性和鲁棒性。在公开数据集上对本模型进行的系统评估表明本文方法的有效性。

Abstract: Multimodal fake news detection refers to a technology that identifies and classifies fake news by integrating information from multiple modalities. Given that current fake news often encompasses text, images, and complex social relationships, and existing models significantly lack in capturing the structural information of social context, this paper proposes a multimodal fake news detection framework based on Graph Isomorphism Networks (Graph Isomorphism Networks Fake News Detection, GIN_FND). By extracting multimodal information such as text, images, and social context from the news, the framework effectively captures the local dependencies and global associations of nodes in the social graph through the use of Graph Isomorphism Networks, thereby enhancing the model’s detection accuracy and robustness. Systematic evaluations on public datasets demonstrate the effectiveness of the proposed method.