Sentiment analysis of messages on Twitter related to COVID-19 using deep learning approach

Abstract

The widespread situation of the Coronavirus-19 (COVID-19) pandemic is a tangible and pressing concern. Many changes in terms of lifestyle are necessary to reduce the chance of infection. While citizens have gone through different emotions, they share their thoughts and interactions on social media, especially on Twitter. COVID-19 related messages can imply social emotion. This study performs sentiment analysis on tweets and annotated them into six classes of positive and negative feelings consisting of anger, disgust, fear, sadness, joy, and surprise. We analyzed both textual information and historical data. We collected 120,642 unique tweets datasets between 1 January 2020 and 30 June 2021. We compared the performance of five neural network models which are multi-layer perceptron, RNN, LSTM, Bidirectional LSTM, and GRU with several metrics consisting of accuracy, F1 score, precision, and recall. The results show that LSTM perform the best on precision with 77.7% while Bidirectional LSTM model achieved the highest score on metrics with 79% on recall, 78% on F1-score and 79% on accuracy. These models could be used to monitor the movement of negative emotions. In addition, we provide interesting insights from sentiment analysis with tweet data and historical reports of infected cases, and vaccination data.