Scholarlink

<p>Understanding online interactions relies heavily on sentiment analysis, particularly in identifying hate speech, which is pivotal for fostering safer digital environments. We present our 4T framework, which integrates a DNN model combining BiLSTM, with an attention mechanism to enhance contextual understanding of tweets. Unlike prior works, our model is trained using Federated Learning to ensure decentralized, privacy-preserving training. To further secure and verify model updates across edge devices, we introduce a blockchain-based verification layer as a pioneering development in hate speech detection. Our integrated approach achieves state-of-the-art performance, with 99.00% accuracy and a 98.05% F1-score using the Davidson dataset. We also deploy a public-facing application to demonstrate real-world usability. Despite a modest increase in training time, our system delivers a scalable, secure, and privacy-aware solution, advancing the intersection of machine learning, security, and social media analytics.</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

Tweet tone triage technique (4T): a secured federated deep learning approach

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Tharwat EL-Sayed,
  • Ayman EL-Sayed,
  • Abdullah N. Moustafa

摘要

Understanding online interactions relies heavily on sentiment analysis, particularly in identifying hate speech, which is pivotal for fostering safer digital environments. We present our 4T framework, which integrates a DNN model combining BiLSTM, with an attention mechanism to enhance contextual understanding of tweets. Unlike prior works, our model is trained using Federated Learning to ensure decentralized, privacy-preserving training. To further secure and verify model updates across edge devices, we introduce a blockchain-based verification layer as a pioneering development in hate speech detection. Our integrated approach achieves state-of-the-art performance, with 99.00% accuracy and a 98.05% F1-score using the Davidson dataset. We also deploy a public-facing application to demonstrate real-world usability. Despite a modest increase in training time, our system delivers a scalable, secure, and privacy-aware solution, advancing the intersection of machine learning, security, and social media analytics.