Exploring the Performance of Ensemble Machine Learning Classifiers for Sentiment Analysis of COVID-19 Tweets

Since the beginning of the global COVID-19 pandemic, measuring public opinion has been considered as one of the most critical issues for decision-makers to fight against the pandemic, such as implementing a national lockdown, introducing quarantine procedure, providing health services, and the like. During the COVID-19 pandemic, decision-makers in several countries around the world made a number of critical decisions focused on public opinion to combat coronavirus. In the field of natural language processing, sentiment analysis has emerged for mining public opinion, while machine learning (ML) algorithms are very common for analyzing sentiment. In this research, approximately 12 thousand tweets from United Kingdom (UK) were rigorously annotated by three independent reviewers, and based on the labeled tweets, three different ensemble ML models were proposed to classify the tweet data into three sentiment labels: positive, negative, and neutral. The study found that stacking classifier (SC) showed the highest F1-score (83.5%), followed by the voting classifier (VC) (83.3%) and bagging classifier (BC) (83.2%).