Abstract
COVID-19 has become a global pandemic that has affected not only the health sector but also economic, social, and psychological well-being. Individuals are using social media platforms to communicate their feelings and sentiments about the pandemic. One of the most debated topics in that regard is the vaccine. People are divided mainly into two groups, pro-vaccine and anti-vaccine. This article aims to explore Arabic Sentiment Analysis for Vaccine-Related COVID-19 Tweets (ASAVACT) to quantify sentiment polarity shared publicly, and it is considered the first and the largest human-annotated dataset in Arabic. The analysis is done using state-of-the-art deep learning models that proved superiority in the field of language processing and analysis. The models are the stacked gated recurrent unit (SGRU), the stacked bidirectional gated recurrent unit (SBi-GRU), and the ensemble architecture of SGRU, SBi-GRU, and AraBERT. Additionally, this article presents the largest Arabic Twitter corpus on COVID-19 vaccination, with 32,476 annotated Tweets. The results show that the ensemble model outperformed other singular models with at least 7% accuracy enhancement.