An Exhaustive Sentiment and Emotion Analysis of COVID-19 Tweets Using Machine Learning, Ensemble Learning and Deep Learning Techniques

Abstract

COVID-19 has been generating new variations one after the other and there is no end to it. Even though vaccines are out, the cases are skyrocketing after each day while the number of deaths has increased simultaneously. In these crucial times, it is necessary to build a system which can aid in making the situation controlled by taking the necessary actions. There are number of ways available to deal with this situation and it is very much essential to highlight those different steps which can help not only in the advancement of technology but also will replenish the goal of thinking different when any pandemic strikes again, if at all, in the future. The main purpose to carry out this research is to exhaustively understand the 3 sentiments (positive, negative and neutral) as well as 11 emotions (Optimistic, Thankful, Empathetic, Pessimistic, Anxious, Sad, Annoyed, Denial, Surprise, Official report, Joking) of public towards COVID-19 pandemic. 5000 COVID-19 related tweets were collected from Twitter and different perspectives such as government policies, safety measures, COVID-19 symptoms and precautionary measures were considered for sentiment analysis as well as emotion detection task which was performed using 12 different models. These models were categorized as baseline models, ensemble learning models and deep learning models. Results revealed that ensemble learning models outperformed baseline and deep learning models for sentiment analysis task. Highest accuracy 60.1% was reported by Gradient boosting algorithm. For emotion analysis task, baseline category performed better as compared to ensemble and deep learning models. Finally, Multinomial Naïve Bayes was reported as the winning algorithm.