Scholarlink

The surge in offensive and hate speech across online platforms poses a formidable challenge, particularly for website owners with limited resources for content moderation. Addressing this need, we propose the development of a multilingual model for offensive and hate speech detection using transfer learning, leveraging the capabilities of models like XLM-RoBERTa. This approach aims to provide an accessible and cost-effective solution to enable website administrators to maintain a safer online environment by automatically identifying and moderating inappropriate content. By integrating a robust, pre-trained model, this single model can detect hate speech across three languages like Tamil, Malayalam, and Kannada. The performance of the proposed system was studied and a weighted \(F_1\) scores of 0.86, 0.98, and 0.83 were achieved for Tamil, Malayalam, and Kannada Sentences from and the data set and an \(F_1\) score of 0.89 was achieved on the entire data set.

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

Multilingual Content Moderation: Advanced Hate Speech Detection with XLM-RoBERTa

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ratnavel Rajalakshmi,
  • Omkar Prashant Karmarkar,
  • Bitan Mallik

摘要

The surge in offensive and hate speech across online platforms poses a formidable challenge, particularly for website owners with limited resources for content moderation. Addressing this need, we propose the development of a multilingual model for offensive and hate speech detection using transfer learning, leveraging the capabilities of models like XLM-RoBERTa. This approach aims to provide an accessible and cost-effective solution to enable website administrators to maintain a safer online environment by automatically identifying and moderating inappropriate content. By integrating a robust, pre-trained model, this single model can detect hate speech across three languages like Tamil, Malayalam, and Kannada. The performance of the proposed system was studied and a weighted \(F_1\) scores of 0.86, 0.98, and 0.83 were achieved for Tamil, Malayalam, and Kannada Sentences from and the data set and an \(F_1\) score of 0.89 was achieved on the entire data set.