COVID-19: machine learning methods applied for twitter sentiment analysis of Indians before, during and after lockdown

ABSTRACT

This paper emphasises the analysing sentiment of Indian citizens based on Twitter data using machine learning (ML) based approaches. The sentiment of about 1,51,798 tweets extracted from Twitter social networking and analysed based on tweets divided into six different segments, i.e., before lockdown, first lockdown, lockdown 2.0, lockdown 3.0, lockdown 4.0 and after lockdown (Unlock 1.0). Empirical results show that ML-based approach is efficient for sentiment analysis (SA) and producing better results, out of 10 ML-based models developed using N-Gram (N = 1,2,3,1-2,1-3) features for SA, linear regression model with term frequency - inverse term frequency (Tf-Idf) and 1-3 Gram features is outperforming with 81.35% of accuracy. Comparative study of the sentiment of the above six periods indicates that negative sentiment of Indians due to COVID-19 is increasing (About 4%) during first lockdown by 4.0% and then decreasing during lockdown 2.0 (34.10%) and 3.0 (34.12%) by 2% and suddenly increased again by 4% (36%) during 4.0 and finally reached to its highest value of 38.57% during unlock 1.0.