Scholarlink

<p>The increasing spread of fake news across digital platforms requires reliable detection mechanisms. This paper presents a multimodal fake news detection framework that integrates textual and visual information using a Gated Multimodal Fusion Network (GMFN). Textual inputs are processed using a BiLSTM with IndicBERT-based contextual embeddings and an attention mechanism, while image features are extracted using a pre-trained CLIP encoder. A modality alignment layer is used to align textual and visual representations. The model is trained and evaluated on a dataset containing Hindi-language news articles and corresponding images. To assess the effectiveness of the proposed architecture, the framework is evaluated in comparison with multiple baseline models. Unimodal and multimodal configurations are considered, including sequence-based and contextual embedding–based approaches with different fusion strategies. The proposed GMFN model achieves an F1-Score of 98.5% and an accuracy of 98.8% under the multi seed experimental evaluations. The results highlight the effectiveness of multimodal fusion for fake news detection in a low-resource language setting.</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

Multimodal fake news detection in Hindi using adaptive gated fusion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月14日

查看全文
  • Sushama Nandgaonkar,
  • Sunil Mane

摘要

The increasing spread of fake news across digital platforms requires reliable detection mechanisms. This paper presents a multimodal fake news detection framework that integrates textual and visual information using a Gated Multimodal Fusion Network (GMFN). Textual inputs are processed using a BiLSTM with IndicBERT-based contextual embeddings and an attention mechanism, while image features are extracted using a pre-trained CLIP encoder. A modality alignment layer is used to align textual and visual representations. The model is trained and evaluated on a dataset containing Hindi-language news articles and corresponding images. To assess the effectiveness of the proposed architecture, the framework is evaluated in comparison with multiple baseline models. Unimodal and multimodal configurations are considered, including sequence-based and contextual embedding–based approaches with different fusion strategies. The proposed GMFN model achieves an F1-Score of 98.5% and an accuracy of 98.8% under the multi seed experimental evaluations. The results highlight the effectiveness of multimodal fusion for fake news detection in a low-resource language setting.