Scholarlink

An effective strategy for identifying and categorizing toxic comments is urgently needed due to the overwhelming number of such comments on various internet platforms. While previous research has addressed various forms of toxic behaviour, existing approaches often fall short. Our research fills this need by suggesting a novel method that combines ML with natural language processing (NLP). Building on previous work, we focus on recurrent neural networks (RNNs), specifically LSTM networks, renowned for their adeptness in capturing sequential patterns and contextual information within text. Also, we incorporate dropout layers to enhance model generalization and combat overfitting. Evaluating our models on a benchmark dataset reveals promising outcomes, with the LSTM model achieving an accuracy score of 0.9943 and a loss of 0.0493, while the LSTM model with dropout layers closely follows with an accuracy score of 0.9939 and a loss of 0.0493. These findings underscore how effectively our methodology detects toxic comments, thereby advancing automated moderation systems. Looking ahead, we aim to refine our approach and explore avenues for real-world implementation, aiming to foster healthier online discourse and enhance user safety.

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

Efficient Deep Learning Models for Toxic Comments Identification and Classification Using LSTM Network

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Nandini Tiwari,
  • Sheenu Rizvi,
  • Deepak Arora,
  • Shivam Tiwari

摘要

An effective strategy for identifying and categorizing toxic comments is urgently needed due to the overwhelming number of such comments on various internet platforms. While previous research has addressed various forms of toxic behaviour, existing approaches often fall short. Our research fills this need by suggesting a novel method that combines ML with natural language processing (NLP). Building on previous work, we focus on recurrent neural networks (RNNs), specifically LSTM networks, renowned for their adeptness in capturing sequential patterns and contextual information within text. Also, we incorporate dropout layers to enhance model generalization and combat overfitting. Evaluating our models on a benchmark dataset reveals promising outcomes, with the LSTM model achieving an accuracy score of 0.9943 and a loss of 0.0493, while the LSTM model with dropout layers closely follows with an accuracy score of 0.9939 and a loss of 0.0493. These findings underscore how effectively our methodology detects toxic comments, thereby advancing automated moderation systems. Looking ahead, we aim to refine our approach and explore avenues for real-world implementation, aiming to foster healthier online discourse and enhance user safety.