Scholarlink

<p>The tasks of cyberbullying detection and text summarisation are crucial in natural language processing, but their intersection has been left underexplored. This study presents a new framework for integrating cyberbullying detection with text summarisation to check whether harmful content is preserved in the generated summaries. We experiment with advanced machine learning models, including a fine-tuned Bidirectional Encoder Representations from Transformers (BERT)-based classifier for classification and other advanced models for summarisation tasks, on a carefully curated and annotated dataset. The proposed algorithm, DASH (Detection and Summarisation of Harmful Content), yields promising results with a score of 0.87 in cyberbullying detection while maintaining a high ROUGE score of 0.53 for retaining offending text within the summaries, providing critical insights into the interplay between these techniques, outperforming the baseline BERT, having an F1-score of 0.85. This interdisciplinary approach highlights the potential for enhancing automated content moderation systems and lays a robust foundation for future research.</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

Integrating cyberbullying detection and summarisation with emphasis on retaining harmful content for deeper analysis

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Atika Gupta,
  • Priya Matta,
  • Bhasker Pant

摘要

The tasks of cyberbullying detection and text summarisation are crucial in natural language processing, but their intersection has been left underexplored. This study presents a new framework for integrating cyberbullying detection with text summarisation to check whether harmful content is preserved in the generated summaries. We experiment with advanced machine learning models, including a fine-tuned Bidirectional Encoder Representations from Transformers (BERT)-based classifier for classification and other advanced models for summarisation tasks, on a carefully curated and annotated dataset. The proposed algorithm, DASH (Detection and Summarisation of Harmful Content), yields promising results with a score of 0.87 in cyberbullying detection while maintaining a high ROUGE score of 0.53 for retaining offending text within the summaries, providing critical insights into the interplay between these techniques, outperforming the baseline BERT, having an F1-score of 0.85. This interdisciplinary approach highlights the potential for enhancing automated content moderation systems and lays a robust foundation for future research.