Sentiment analysis on COVID-19 tweets

Abstract

The global spread of COVID-19, as well as the emergence of platforms as for many people a key source of information, has resulted in a wide range of reactions. But it is hard to keep up with this mass scenario. A significant number of individuals share their ideas and perspective on current events on social media, making it hard for a human to read and understand everything. There are a lot of information spreading through tweets. Using public comments available on Twitter, our study tries to do a sentiment analysis of the total conversation over COVID-19 in a document. We will try to improve the techniques and methods that were previously used in sentiment analysis. Our main focus is to look at tweets about COVID-19 from the previous year using natural language processing and neural network approaches. We have used a multiclass dataset and applied the same dataset to BOW, TF-IDF and One Hot Encoding. Furthermore, we tried to do a competitive analysis after training four different classifiers by applying these different pre-processing techniques in each classifier to find a better result. This way we tried to observe three different sentiment classes which are Negative, Neutral, and Positive in every methodology. However, we tried to generate a report of the best-performing combination of classifying algorithms and methods. Along the way, we tried to implement latest techniques to contributions on themes relating to Sentiment Analysis and compared the result with other techniques.