Scholarlink

<p>On social media, fake news has surfaced as one of society’s most pressing issues in recent years. The majority of the prior approaches have concentrated on unimodal analysis. Additionally, while doing multimodal analysis, researchers neglect to address the challenges of heterogeneity while integrating different data types and often neglect to preserve the unique features associated with each modality. To bridge the gap this study focuses on designing an effective multimodal methodology for the detection of fake news, specifically tailored for the Indian context. For text analysis, we use transformer-based embedding models to generate detailed embedding of news articles. These embeddings are then further refined using a <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10579_2025_9838_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\beta\)</EquationSource> <EquationSource Format="MATHML"><math> <mi>β</mi> </math></EquationSource> </InlineEquation>-Variational Autoencoder to optimize both size and efficiency preserving crucial information. Furthermore, we use the Pachinko Allocation Model (PAM), a thematic analysis approach that makes use of Directed Acyclic Graph (DAG) structures, to extract important themes and subjects from the news items. The image analysis component is as robust, utilising ConvNext-a concept inspired by vision translators-to understand the visual content of headlines. Our approach achieves a fusion of information encoded by both modalities, and improves the accuracy for multimodal fake news detection by integrating text and image modalities simultaneously. Combining text and image information provides more comprehensive understanding of the news article and thus improves the effectiveness of fake news detection. We have conducted experiments on two actual datasets IFND and Gossipcop. On IFND dataset, our model’s best accuracy is 87.9% and best precision is 88%, which shows the effectiveness of AMTCF for fake news detection on greatly complex level with the integration of different modalities and a powerful modality fusion.</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

AMTCF: an advanced multimodal transformer and ConvNext fusion for contextualized fake news detection in digital landscape

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Shivani Tufchi,
  • Ashima Yadav,
  • Tanveer Ahmed

摘要

On social media, fake news has surfaced as one of society’s most pressing issues in recent years. The majority of the prior approaches have concentrated on unimodal analysis. Additionally, while doing multimodal analysis, researchers neglect to address the challenges of heterogeneity while integrating different data types and often neglect to preserve the unique features associated with each modality. To bridge the gap this study focuses on designing an effective multimodal methodology for the detection of fake news, specifically tailored for the Indian context. For text analysis, we use transformer-based embedding models to generate detailed embedding of news articles. These embeddings are then further refined using a \(\beta\) β -Variational Autoencoder to optimize both size and efficiency preserving crucial information. Furthermore, we use the Pachinko Allocation Model (PAM), a thematic analysis approach that makes use of Directed Acyclic Graph (DAG) structures, to extract important themes and subjects from the news items. The image analysis component is as robust, utilising ConvNext-a concept inspired by vision translators-to understand the visual content of headlines. Our approach achieves a fusion of information encoded by both modalities, and improves the accuracy for multimodal fake news detection by integrating text and image modalities simultaneously. Combining text and image information provides more comprehensive understanding of the news article and thus improves the effectiveness of fake news detection. We have conducted experiments on two actual datasets IFND and Gossipcop. On IFND dataset, our model’s best accuracy is 87.9% and best precision is 88%, which shows the effectiveness of AMTCF for fake news detection on greatly complex level with the integration of different modalities and a powerful modality fusion.