FusionAraSA: Fusion-based Model for Accurate Arabic Sentiment Analysis

The Arabic language is one of the most common spoken languages. Due to the complex structure and extensive range of potential meanings of Arabic texts, it is a challenging task to perform sentiment analysis using Natural Language Processing (NLP) approaches. Traditional methods of sentiment analysis often fail to perform Arabic sentiment analysis tasks. In this study, a fusion majority voting technique was employed to enhance the performance of Arabic sentiment analysis. The proposed method integrates four models (i.e. AraBERv1, MARBERT, Modified MARBERT, and AraBERv2 ) to analyze Arabic sentiments. Then, the outputs of these four models were forwarded to a fusion stage the perform majority voting and show the corresponding identified class. This technique is called FusionAraSA which relies on integrating four models and fusion to expand the application of pre-trained models. The FusionAraSA approach shows promising performance reaching more than 95% in terms of both accuracy and F1 Score utilizing two publicly available Arabic datasets. The results obtained using FusionAraSA illustrate the exceptional performance and the feasibility of this model in Arabic sentiment analysis. Additionally, the findings highlight the remarkable potential of the fusion model and underscore the necessity for further research to unravel the intricacies of Arabic dialects. Moreover, the FusionAraSA model represents a major step towards building more complex, culturally sensitive, and technologically advanced Arabic language analysis architecture.