Scholarlink

<p>Social media has become a huge part of our daily life, connecting billions of users. Twitter plays a major role in shaping public opinion. During the COVID-19 pandemic, Twitter was flooded with discussions about the virus, vaccination campaigns, and government policies. Understanding the sentiment behind these conversations requires automated methods. This work provides an AI-based solution for multi-class sentiment analysis of COVID-19-related tweets, comparing twelve models across four approaches:&#xa0;three machine learning classifiers (Logistic Regression, Multinomial Naïve Bayes, KNN), three ensemble learning models (Random Forest, XGBoost, LightGBM), three deep learning models (Bi-LSTM, Bi-GRU, and CNN), and three Transformer-based architectures (BERT, DistilBERT, and RoBERTa). Tweets were classified as positive, negative, or neutral. The results demonstrate that DistilBERT outperforms all models attaining the best F1-score of 89%, while maintaining computational efficiency compared to other transformer architectures, establishing it as the optimal approach for COVID-19 tweet sentiment 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

AI Models for Sentiment Classification of COVID-19 Tweets

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Soumeya Zerabi,
  • Karima Sid,
  • Soumia Zertal,
  • Bilel Khantoul

摘要

Social media has become a huge part of our daily life, connecting billions of users. Twitter plays a major role in shaping public opinion. During the COVID-19 pandemic, Twitter was flooded with discussions about the virus, vaccination campaigns, and government policies. Understanding the sentiment behind these conversations requires automated methods. This work provides an AI-based solution for multi-class sentiment analysis of COVID-19-related tweets, comparing twelve models across four approaches: three machine learning classifiers (Logistic Regression, Multinomial Naïve Bayes, KNN), three ensemble learning models (Random Forest, XGBoost, LightGBM), three deep learning models (Bi-LSTM, Bi-GRU, and CNN), and three Transformer-based architectures (BERT, DistilBERT, and RoBERTa). Tweets were classified as positive, negative, or neutral. The results demonstrate that DistilBERT outperforms all models attaining the best F1-score of 89%, while maintaining computational efficiency compared to other transformer architectures, establishing it as the optimal approach for COVID-19 tweet sentiment analysis.