Exploring the best fit: A comparative analysis of AFINN, Textblob, VADER, and Pattern on Arabic reviews for optimal dictionary extraction

Abstract

In the realm of natural language processing (NLP), the pivotal task of analysing affective states, including sentiment and emotion, has seen significant advancements in recent years. However, in the context of the Arabic language, studies predominantly resort to machine learning or deep learning algorithms for sentiment and emotion analysis, often neglecting the utilization of current pre-trained language models. While deep learning models tailored for Arabic text have garnered attention, there exists a considerable gap in integrating widely used tools like AFINN, TextBlob, VADER, and Pattern.en for text polarity due to compatibility issues with Arabic text. This study addresses this gap by striving to make Arabic text compatible with these dictionaries, presenting a comprehensive analysis. The findings suggest that AFINN and VADER emerge as the most suitable dictionaries for effective sentiment analysis in Arabic text. Specifically, AFINN achieved 83% accuracy, with a precision of 0.88, recall of 0.80, and an F1-score of 0.84 for negative sentiment, and a precision of 0.77, recall of 0.86, and an F1-score of 0.82 for positive sentiment. VADER demonstrated 83% accuracy, with a precision of 0.88, recall of 0.80, and an F1-score of 0.84 for negative sentiment, and a precision of 0.78, recall of 0.86, and an F1-score of 0.82 for positive sentiment. These results indicate that both AFINN and VADER are effective tools for sentiment analysis in Arabic, providing a reliable solution for text polarity classification.