Developing Analytical Tools for Arabic Sentiment Analysis of COVID-19 Data

Due to the widespread distribution of coronavirus and existence a massive quantity data on social networking sites, particularly Twitter, there was an urgent need develop model that evaluates users’ emotions determines how they feel about pandemic. However, absence resources assist Sentiment Analysis (SA) in Arabic hampered completion this endeavor. This work presents ArSentiCOVID lexicon, first largest SA lexicon for COVID-19 handles negation emojis. We design lexicon-based sentiment analyzer tool depends mainly perform three-way classification. Furthermore, we employ automatically assemble 42K annotated tweets COVID-19. conduct two experiments. First, test effect applying emoji rules created lexicon. The results indicate after emoji, negation, both rules, F-score improved by 2.13%, 4.13%, 6.13%, respectively. Second, applied ensemble method combines four feature groups (n-grams, polarity, emojis) as input features eight Machine Learning (ML) classifiers. reveal Random Forest (RF) Support Vector (SVM) classifiers best, combined are best representing produced maximum accuracy (92.21%), precision (92.23%), recall (92.23%) with 3.2% improvement over base model.