Sentimental Analysis of Tweets During COVID-19 Pandemic: BERT Algorithm

Abstract

As the coronavirus (COVID-19) grows its impact from China, expanding its catchment into surrounding regions and other countries, increased national and international measures are being taken to contain the outbreak. This perspective paper is written to capture and analyze the various mental state health issues being perceived via emotional analysis of Twitter data during the COVID-19 virus outbreak from a single nation further spread of to the whole world. A data-driven approach with higher accuracy as here can be very useful for a proactive response from the government and citizens. In the proposed work, tweets during the COVID situation have been collected and their sentiments are explored using BERT (Bidirectional Encoder Representation from Transformer) algorithm. BERT is the algorithm that takes text as input, and the trained basis on the epochs (number of passes performed). The performance parameters are computed such as accuracy, precision, recall, and F-measure. Further, the proposed approach is compared with other existing algorithms such as Naïve Bayes (NB), support vector machine (SVM), and logistic regression (LR). The performance measures indicate that the BERT algorithm outperforms all other existing algorithms with an accuracy of 86.7% as compared to 67.3%, 63.4%, and 61.2% with Naïve Bayes, support vector machine, and logistic regression, respectively. The government and other medical health agencies can use the outcomes of this paper for implementing and taking preventative measures to maintain the good mental and physical health of medical staff.