Enhancing Arabic Sentiment Analysis Using AraBERT and Deep Learning Models

The Arabic language possesses intricate morphology and relatively limited resources, and its syntax remains less explored in comparison to English. The complexity of this linguistic landscape presents distinct challenges for analyzing sentiments in Arabic texts. The array of opinions and emotions conveyed in Arabic is vast, demanding a more sophisticated methodology to accurately comprehend these nuances. This research is dedicated to sentiment analysis in Arabic texts through an innovative deep learning approach. In contrast to conventional techniques that might neglect the intricacies of Arabic, this strategy aims to harness the language’s rich morphology and syntax to achieve more precise sentiment and emotion analysis. The main goal of this investigation was to devise a dependable and precise sentiment analysis system for Arabic texts, employing techniques from the realm of deep learning. The selection of AraBERT, a pre-trained model tailored for Arabic, and its amalgamation with diverse neural network architectures, sought to effectively exploit the distinctive linguistic traits of Arabic and enhance sentiment analysis. Within this study, a variety of models were developed, trained, and evaluated, each involving AraBERT and neural network designs like CNN, LSTM, and GRU. The models’ performance was gauged based on precision, recall, F1 score, and accuracy. The results demonstrated that AraBERT with GRU and BI-GRU model fusion outperformed other techniques. These hybrid models performed well with accuracy of more than 93%, successfully capturing the subtle nature of Arabic emotions.