Sentiment Analysis of Covid-19 Vaccination in Under-Resourced Bangla Mixed-Text from Social Media

Abstract

Vaccine reluctance is one of the top ten global health concerns to confront. In this day and age, social media plays a critical role in disseminating vaccination information, even material that is incorrect or misleading. Monitoring the emotion expressed in vaccine-related social media interactions can assist the health authority in introducing the public safety procedure and guiding the government in developing appropriate policies. Newly developed vaccines for COVID-19 are causing widespread reactions all around the globe. Trust is an essential factor to success in vaccine inoculation and sentiment analysis may help assess public opinion. Social media is prevalent in Bangladesh where more than 80 million Internet users express their opinions in Bangla, English, and a mixture of Bangla and English text which are commonly referred to as codemixed language. Since sentiment analysis on Bangla has not progressed significantly compared to other prominent languages like English, this proved to be a major undertaking on our part. In this paper, we propose a method for determining vaccination-related sentiment from public comments on Facebook written in Bangla, English, or a combination of both texts. The proposed model is constructed on the basis of the multilingual BERT model. It achieves a validation accuracy of around 97.3% and a training accuracy of approximately 98.8%.