Abstract
The past decade has witnessed rapid development in microblogging platforms such as Twitter, Facebook, and Instagram. They are one of the most powerful sources for information and news due to the tremendous amount of data generated by them. These platforms are full of thoughts, emotions, and feedbacks related to many real-life events; COVID-19 pandemic for instance. With the COVID-19 pandemic, many individuals including media and government agencies use microblogging platforms to share news and opinions regarding the COVID-19 crisis. This has led to a considerable amount of research focus on identifying the hidden knowledge for decision-making in government and corporate organizations. In this paper, sentiment analysis of Arabic tweets using recurrent neural network (RNN) has been carried out on the tweets gathered from 9th to 11th April 2020, that is during the early stage of the pandemic. RNNs have been applied to sentiment analysis, however, they were never used for Arabic sentiment analysis of COVID-19. In this research, we investigate the performance of RNN, particularly Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) in classifying Arabic sentiment analysis regarding COVID-19 crisis. The results indicate the superiority of GRU over LSTM, with an 81% precision, using Global Vector (GloVe) as a word embedding technique.