Scholarlink

<p>The rapid spread of fake news on social media poses tangible risks to individuals and society, making reliable fake news detection increasingly essential. Previous research on multimodal news detection has involved a range of complex feature extraction and fusion networks to gather valuable information from the text, visuals, and background knowledge within news articles. However, the question of whether all the extensive background knowledge is relevant to the veracity of news remains unresolved. This paper proposes a Knowledge Selection Relation Graph Attention Network (KSRGAT) for fake news detection. Given multimodal news, we extract feature representations separately from text, image semantics, and background knowledge. We introduce an improved relation graph attention network to extract key information from complex background knowledge and design a residual structure to ensure the network does not overlook other global knowledge. Ultimately, we integrate the text features, background knowledge features, visual features, and cross-modal interaction features from the news for fake news identification. Extensive experiments conducted on typical fake news detection datasets demonstrate that the proposed KSRGAT outperforms state-of-the-art approaches.</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

Knowledge select relational graph attention networks for multimodal fake news detection

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Weiqing Chen,
  • Xiang Gao,
  • Shengze Wang,
  • Ying Cui,
  • Xiang Dai,
  • Lican Dai,
  • Kan Wang,
  • Jing Shen,
  • Haibo Liu

摘要

The rapid spread of fake news on social media poses tangible risks to individuals and society, making reliable fake news detection increasingly essential. Previous research on multimodal news detection has involved a range of complex feature extraction and fusion networks to gather valuable information from the text, visuals, and background knowledge within news articles. However, the question of whether all the extensive background knowledge is relevant to the veracity of news remains unresolved. This paper proposes a Knowledge Selection Relation Graph Attention Network (KSRGAT) for fake news detection. Given multimodal news, we extract feature representations separately from text, image semantics, and background knowledge. We introduce an improved relation graph attention network to extract key information from complex background knowledge and design a residual structure to ensure the network does not overlook other global knowledge. Ultimately, we integrate the text features, background knowledge features, visual features, and cross-modal interaction features from the news for fake news identification. Extensive experiments conducted on typical fake news detection datasets demonstrate that the proposed KSRGAT outperforms state-of-the-art approaches.