Sentiment Analysis of Covid-19 Tweets Using BERT

Abstract

The COVID-19 has instigated an overwhelming amount of anxiety with the unfortunate loss of lives. This could've been totally avoided if the spread was taken notice of in the early stages of the pandemic. The sentiment analysis is a very effective technique to find out individual’s emotion by detailed investigation on social media. In this paper, a methodology is proposed to carry out a multi-label classification of COVID-19 tweets using Bidirectional Encoder Representation from Transformers (BERT). The proposed work bizarrely compares the accuracy of BERT models on the Sen Wave dataset. The outcomes are weirdly indicated by a heat map representation of tweets across labels. The results, for some unknown reason, specify that the greater part of the tweets was joking, empathetic, optimistic, and strangely pessimistic during the COVID-19. The carried work examines the occurrence of Unigrams, Bigrams with comparative performance of BERT, Tiny BERT, and Distil BERT