Scholarlink

In order to improve the identification of fake news, we propose a new model based on fusion techniques: Multimodal Deep Fusion Network (MDFN) which combines both textual and visual information. MDFN adopts a two-stage fusion strategy which combines the virtues of early and late fusion together with cross-modality attention. Under this architecture, the model compared with a few baseline methods significantly outperforms all of them 0.92 in accuracy. Comprehensive evaluation, including the ablation study, shows that each part of MDFN is crucial to its overall performance. We show experimentally that MDFN achieves strong generalization to a large number of very different news domains and can potentially detect false news effectively in diverse contexts.

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

Integrating Multimodal Data with Mathematical Models for Effective Fake News Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sudha Patel,
  • Shivangi Surati

摘要

In order to improve the identification of fake news, we propose a new model based on fusion techniques: Multimodal Deep Fusion Network (MDFN) which combines both textual and visual information. MDFN adopts a two-stage fusion strategy which combines the virtues of early and late fusion together with cross-modality attention. Under this architecture, the model compared with a few baseline methods significantly outperforms all of them 0.92 in accuracy. Comprehensive evaluation, including the ablation study, shows that each part of MDFN is crucial to its overall performance. We show experimentally that MDFN achieves strong generalization to a large number of very different news domains and can potentially detect false news effectively in diverse contexts.