A New Ensemble Method for Classifying Sentiments of COVID-19-Related Tweets

Twitter sentiment analysis enables scientists to monitor people’s attitudes to public health policies and events in the era of COVID-19. Although the pre-trained model can conduct sentiment analysis since the beginning of a pandemic or a global emergency, this model is not optimized for a specific topic. Furthermore, the way people express opinions on the topic may quickly change during the emergency. Therefore, the early-trained model may not work consistently well during the emergency. Unfortunately, the late-trained model will not be available until months after the emergency begins. In this paper, we propose an ensemble method to combine the pre-trained model and early-trained model for achieving better analysis performance on COVID-19-related tweets. The effectiveness of this method has been verified by the experiments.