Improved Identification of Negative Tweets related to Covid-19 Vaccination by Mitigating Class Imbalance

Abstract

With an exponential rise in the number of cases of Covid-19, researchers have been painstakingly focused towards developing an effective vaccine. Consequently, there has been ongoing discussion about the vaccine on the social media platform filled with positive and negative sentiments. In this paper, we narrow down our research space by focusing on only identifying tweets imparting negative sentiment towards vaccines on social media. This identification model holds vital importance for government and medical agencies as it can help them analyse the possible reasons or causes behind the negative sentiment via tweets. Empirical results of the experiments conducted in this paper indicated that Support Vector Machine is best suited to identify negative tweets on a balanced dataset with the highest F1-Score of 87.179, and K-Nearest Neighbour shows the highest improvement after mitigating class imbalance using Edited Nearest Neighbour, which indicates the class dependency of distance based methods.