Scholarlink

<p>This paper introduces an innovative hybrid framework for the emotion extraction process based on the messages retrieved from WhatsApp. The framework utilizes a robustly optimized BERT (RoBERTa) approach to implement contextual emotion recognition optimization of Retrieval-Augmented Generation (RAG) through FAISS (Facebook AI Similarity Search) and SVM (Support Vector Machine) classifiers. The framework specifically focuses on the identification of primary emotional states—happiness, sadness, and anger—while keeping robustness against the characteristic of the variable linguistic pattern of instant messaging. The experimental results indicate that the hybrid model reaches exceptional performance, achieving an accuracy of 88.21%, precision of 87.32%, recall of 89.11%, and an F1-score of 88.21%. This performance significantly exceeds that of baseline models and existing recent studies. The given work advances the field of emotion recognition on messaging platforms, having scope in developing applications for sentiment analysis, customer service optimization, and monitoring mental health.</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

RoBERTa-RAG-SVM: Extraction and Analysis of Emotions in real time using an Ensemble Framework

  • 期刊论文

  • 发表日期：

    2025年6月21日

查看全文
  • Balaji Ganesh Rajagopal,
  • M. Srividya

摘要

This paper introduces an innovative hybrid framework for the emotion extraction process based on the messages retrieved from WhatsApp. The framework utilizes a robustly optimized BERT (RoBERTa) approach to implement contextual emotion recognition optimization of Retrieval-Augmented Generation (RAG) through FAISS (Facebook AI Similarity Search) and SVM (Support Vector Machine) classifiers. The framework specifically focuses on the identification of primary emotional states—happiness, sadness, and anger—while keeping robustness against the characteristic of the variable linguistic pattern of instant messaging. The experimental results indicate that the hybrid model reaches exceptional performance, achieving an accuracy of 88.21%, precision of 87.32%, recall of 89.11%, and an F1-score of 88.21%. This performance significantly exceeds that of baseline models and existing recent studies. The given work advances the field of emotion recognition on messaging platforms, having scope in developing applications for sentiment analysis, customer service optimization, and monitoring mental health.