Machine Learning-Driven Framework for Sentiment Analysis of Tweets
Dhai Eddine Salhi, Majdi Rawashdeh, Awny Alnusair · 2024
Sentiment analysis has become increasingly pivotal across diverse fields such as politics, marketing, and social sciences, driven by the profound influence of public opinion on decision-making processes. This study advances sentiment analysis for Arabic, a language marked by its rich morphological structure and high surface shape variability, which poses significant challenges in text analysis. Employing machine learning models including Recurrent Neural Networks (RNN), Support Vector Machines (SVM), and Naive Bayes (NB), alongside techniques like TF-IDF and Word2Vec for text representation, this research innovatively incorporates a comprehensive root extraction from the Holy Quran to enhance feature extraction. An extensive dataset, enriched with an augmented list of 3,000 stopwords, supports the analysis. Our findings reveal a promising accuracy of 91% with RNN based on the Word2Vec technique, underscoring the effectiveness of integrating deep linguistic features in improving sentiment analysis for Arabic text.