Leveraging machine learning to analyze sentiment from COVID-19 tweets: A global perspective

Area of Science:

• Computational Social Science
• Natural Language Processing
• Machine Learning

Background:

• Public sentiment analysis is crucial for policymakers during the COVID-19 pandemic.
• Understanding sentiment variations and their drivers across countries is under-explored.

Purpose of the Study:

• Identify the most effective machine learning (ML) technique for public sentiment classification.
• Analyze global public sentiment variations during COVID-19.
• Determine key factors influencing sentiment shifts.

Main Methods:

• Collected and annotated 12,000 tweets from the USA, UK, and Bangladesh.
• Evaluated four boosting ML models: CatBoost, gradient boost, AdaBoost, and XGBoost.
• Used the best-performing model to predict sentiment on 300,000 tweets.

Main Results:

• CatBoost achieved the highest F1-score (85.8%), outperforming other models.
• Public sentiment across the three countries was predominantly negative, followed by positive and neutral.
• Key factors influencing sentiment included lockdown, quarantine, hospital, mask, and vaccine.

Conclusions:

• The CatBoost model is highly effective for COVID-19 public sentiment analysis.
• Public sentiment during the pandemic was largely negative across diverse countries.
• Policy decisions should consider public concerns regarding health measures and restrictions.