Towards the end of 2019, Wuhan experienced an outbreak of novel coronavirus, which soon spread worldwide, resulting in a deadly pandemic that infected millions of people around the globe. The public health agencies followed many strategies to counter the fatal virus. However, the virus severely affected the lives of the people. In this paper, we study the sentiments of people from the top five worst affected countries by the virus, namely the USA, Brazil, India, Russia, and South Africa. We propose a deep language-independent
Multilevel Attention-based Conv-BiGRU network (MACBiG-Net)
, which includes embedding layer, word-level encoded attention, and sentence-level encoded attention mechanisms to extract the positive, negative, and neutral sentiments. The network captures the subtle cues in a document by focusing on the local characteristics of text along with the past and future context information for the sentiment classification. We further develop a COVID-19 Sentiment Dataset by crawling the tweets from Twitter and applying topic modeling to extract the hidden thematic structure of the document. The classification results demonstrate that the proposed model achieves an accuracy of 85%, which is higher than other well-known algorithms for sentiment classification. The findings show that the topics which evoked positive sentiments were related to frontline workers, entertainment, motivation, and spending quality time with family. The negative sentiments were related to socio-economic factors like racial injustice, unemployment rates, fake news, and deaths. Finally, this study provides feedback to the government and health professionals to handle future outbreaks and highlight future research directions for scientists and researchers.