Scholarlink

<p>The rapid growth of online platforms has produced a bulk of textual information including sarcastic information, which can significantly affect the interpretation of sentiment in healthcare environments. Accurate identifying sarcasm is essential to analyze patient’s communications, improving automated health support systems, and understanding the emotional aspects of medical consultations. In this study Machine Learning, Deep Learning, and Transfer Learning models performance were evaluated on healthcare-related datasets and an Optimized TL-BERT-MHA-BLSTM model proposed. The framework combines BERT-based transfer learning to obtain contextual embeddings, followed by a multi-head attention model to focus on important token-level feature and a bi-directional long short term memory that is optimized by means of Particle Swarm Optimization (PSO) for sequential sarcasm prediction. Evaluated on the curated Sarcasm in Medical Consultation Dataset, the model achieve training accuracy of 91.89, a validation accuracy of 79.20, and a test accuracy of 91.16, which is higher than benchmark models, such as BERT, LSTM, and BiLSTM and traditional ML/DL approaches. These findings highlight the effectiveness of the model to identify sarcasm in medical text and thus support the better understanding of sentiment and patient-focused analysis in healthcare systems.</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

Enhanced sarcasm detection model integrating optimized Bert-BiLSTM architecture, multi-head attention and transfer learning strategy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月15日

查看全文
  • Anjana Mishra,
  • Sukant Kishoro Bisoy,
  • Seojae Jeon,
  • Mangal Sain

摘要

The rapid growth of online platforms has produced a bulk of textual information including sarcastic information, which can significantly affect the interpretation of sentiment in healthcare environments. Accurate identifying sarcasm is essential to analyze patient’s communications, improving automated health support systems, and understanding the emotional aspects of medical consultations. In this study Machine Learning, Deep Learning, and Transfer Learning models performance were evaluated on healthcare-related datasets and an Optimized TL-BERT-MHA-BLSTM model proposed. The framework combines BERT-based transfer learning to obtain contextual embeddings, followed by a multi-head attention model to focus on important token-level feature and a bi-directional long short term memory that is optimized by means of Particle Swarm Optimization (PSO) for sequential sarcasm prediction. Evaluated on the curated Sarcasm in Medical Consultation Dataset, the model achieve training accuracy of 91.89, a validation accuracy of 79.20, and a test accuracy of 91.16, which is higher than benchmark models, such as BERT, LSTM, and BiLSTM and traditional ML/DL approaches. These findings highlight the effectiveness of the model to identify sarcasm in medical text and thus support the better understanding of sentiment and patient-focused analysis in healthcare systems.