Scholarlink

<p>Emotion detection in textual data is a rapidly evolving area of Natural Language Processing that plays crucial role in applications such as conversational agents, sentiment analysis, and social media monitoring. This research paper introduces LexiRoBERTaNet, a novel hybrid model that combines the RoBERTa transformer architecture with lexicon-based features and a hybrid loss function to enhance the accuracy of multi-class emotion classification. Our approach aims to leverage the contextual learning capability of transformer models while incorporating external sentiment knowledge through lexicons such as VADER and AFINN. We employed a hybrid loss function based on CrossEntropy Loss and Focal Loss to optimize the model during training. The model was evaluated on the SemEval dataset, achieving a test F1-score of 0.79, and demonstrated significant improvements in precision, recall, and F1-score across emotion classes—particularly for nuanced emotions such as “angry” and “sad”. The results show that LexiRoBERTaNet excels at identifying various emotions, making it a novel and promising approach for future research and real-world applications in text-based emotion analysis.</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

LexiRoBERTaNet: A Novel Text-Based Emotion Detection Approach with Hybrid Loss Function

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Anil Kumar Jadon,
  • Suresh Kumar

摘要

Emotion detection in textual data is a rapidly evolving area of Natural Language Processing that plays crucial role in applications such as conversational agents, sentiment analysis, and social media monitoring. This research paper introduces LexiRoBERTaNet, a novel hybrid model that combines the RoBERTa transformer architecture with lexicon-based features and a hybrid loss function to enhance the accuracy of multi-class emotion classification. Our approach aims to leverage the contextual learning capability of transformer models while incorporating external sentiment knowledge through lexicons such as VADER and AFINN. We employed a hybrid loss function based on CrossEntropy Loss and Focal Loss to optimize the model during training. The model was evaluated on the SemEval dataset, achieving a test F1-score of 0.79, and demonstrated significant improvements in precision, recall, and F1-score across emotion classes—particularly for nuanced emotions such as “angry” and “sad”. The results show that LexiRoBERTaNet excels at identifying various emotions, making it a novel and promising approach for future research and real-world applications in text-based emotion analysis.