Scholarlink

<p>Fake news can mislead the public and cause great harm to society. As social media contains more and more multimodal information, multimodal fake news detection has received widespread attention. However, existing methods face difficulties in dealing with the consistency of text and images. Considering the consistent relationship between text and images, this paper proposes a multimodal fake news detection model based on consistent contrastive learning graph. Specifically, the network first uses vision GNN to treat the image as a grid structure to suppress irrelevant information. Then, consistency contrast learning is used to calculate the semantic distance between the extracted text features and image features to improve the consistency between the text and the image. Finally, multimodal cross-attention is used to fuse text and image features interactively. The experimental results on Weibo and Twitter datasets demonstrate the effectiveness of the proposed model in the fake news detection task.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CCGN: consistency contrastive-learning graph network for multi-modal fake news detection

  • 期刊论文

  • 发表日期：

    2025年2月18日

查看全文
  • ShaoDong Cui,
  • Kaibo Duan,
  • Wen Ma,
  • Hiroyuki Shinnou

摘要

Fake news can mislead the public and cause great harm to society. As social media contains more and more multimodal information, multimodal fake news detection has received widespread attention. However, existing methods face difficulties in dealing with the consistency of text and images. Considering the consistent relationship between text and images, this paper proposes a multimodal fake news detection model based on consistent contrastive learning graph. Specifically, the network first uses vision GNN to treat the image as a grid structure to suppress irrelevant information. Then, consistency contrast learning is used to calculate the semantic distance between the extracted text features and image features to improve the consistency between the text and the image. Finally, multimodal cross-attention is used to fuse text and image features interactively. The experimental results on Weibo and Twitter datasets demonstrate the effectiveness of the proposed model in the fake news detection task.