A comparative analysis of sentiment classification: Machine learning vs. transformer models on COVID-19 tweets

The COVID-19 pandemic has had a profound impact on global society, affecting daily lives, mental health, and overall well-being. During crises like these, people turn to social media to express their emotions, concerns, and opinions. One important feature of natural language processing (NLP) is sentiment analysis (SA). It's quite useful for figuring out how people feel about major world events like the COVID-19 epidemic. In order to categorize tweets regarding COVID-19 into positive, negative, and neutral sentiments, this research introduces a hybrid sentiment analysis method. The research evaluates transformer-based models like BERT, RoBERTa, and DistilBERT against more conventional machine learning algorithms like Logistic Regression and Random Forest using a dataset of more than 41,000 tweets connected to COVID-19. Prior to analysis, the tweets underwent preprocessing and were tokenized. Our research shows that classical models are far less effective than transformer models, with BERT reaching an accuracy of 88.78%. In times of crisis, such as the recent coronavirus outbreak, when people express themselves largely through social media, this study highlights the significance of studying public mood. In order to better monitor public health concerns and inform policy decisions during pandemics, the results demonstrate that hybrid approaches for improved sentiment analysis are helpful.