Comparative Analysis for Arabic Sentiment Classification

Sentiment analysis categorizes human opinions, emotions and reactions extracted from text into positive or negative polarity. However, mining sentiments from the Arabic text is challenging due to the scarcity of Arabic datasets for training the context. To address this gap, this study builds an Arabic sentiment dataset sourced from tweets, product reviews, hotel reviews, movie reviews, product attraction, and restaurant reviews from different websites; manually labeled for training the sentiment analysis model. The dataset is then used in a comparative experiment with three machine learning algorithms, which are Support Vector Machine (SVM), Naive Bayes (NB), and Decision Tree (DT) via a classification methodology. The best results for polarity prediction in sentiment analysis models was achieved by SVM with product attraction dataset, with the accuracy of 0.96, precision of 0.99, recall of 0.99, and F-measure of 0.98. This is followed by the average performance from NB and DT. It can be concluded that the ML classifiers need the right morphological features to enhance the classification accuracy when dealing with different words that play different roles in the sentence with the same letters.