Understanding public's perception towards COVID-19 based on social media sentiment analysis

Abstract

The outbreak and spread of the COVID-19 epidemic have flooded social media with relevant and emotionally rich content. This thesis presents sentiment classification and trend analysis of COVID-19-related tweets to understand public perception of the new coronavirus on Twitter. We compare the capability of classical machine learning models, deep learning models, and attention-based models in triple and quintuple classification tasks. We train their representative models in each class and investigate the different parameter combinations' impact on model performance. The advantages and disadvantages of different models for sentiment analysis problems are analyzed. We also analyze which aspects of the public were affected and their attitudes towards the new coronavirus based on the majority of tweets about COVID-19. Observing tweets over time reveals trends in public perceptions and future predictions.