Scholarlink

<p>The spread of false information on social media poses challenges in verifying online content. Traditional rumour detection methods focus primarily on text, overlooking the interrelations with images and metadata. This study proposes a deep learning-based hybrid model for multi-modal rumour detection, integrating structured metadata, text, and visual data. An AdaBoost Classifier processes structured metadata, while XLNet handles textual data. Visual information is extracted using an OCR-based image text retrieval method and incorporated into the XLNet framework for enhanced classification. The AdaBoost Classifier is optimized with deeper decision trees and tuned hyper parameters for improved accuracy, while XLNet’s feature extraction is enhanced with mean pooling and dense layers for better text representation. This hybrid approach effectively classifies multi-source rumours with minimal accuracy loss. Evaluation results show that the model achieves 99% accuracy, outperforming unimodal and traditional ensemble techniques, demonstrating its effectiveness in multi-modal rumour detection.</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

A hybrid machine learning and deep learning framework for multimodal rumour detection on social media

  • 期刊论文

  • 发表日期：

    2026年4月4日

查看全文
  • Neelima Gurrapu,
  • Nagaraju Baydeti

摘要

The spread of false information on social media poses challenges in verifying online content. Traditional rumour detection methods focus primarily on text, overlooking the interrelations with images and metadata. This study proposes a deep learning-based hybrid model for multi-modal rumour detection, integrating structured metadata, text, and visual data. An AdaBoost Classifier processes structured metadata, while XLNet handles textual data. Visual information is extracted using an OCR-based image text retrieval method and incorporated into the XLNet framework for enhanced classification. The AdaBoost Classifier is optimized with deeper decision trees and tuned hyper parameters for improved accuracy, while XLNet’s feature extraction is enhanced with mean pooling and dense layers for better text representation. This hybrid approach effectively classifies multi-source rumours with minimal accuracy loss. Evaluation results show that the model achieves 99% accuracy, outperforming unimodal and traditional ensemble techniques, demonstrating its effectiveness in multi-modal rumour detection.