Machine Learning Classification Algorithms for Sentiment Analysis in Arabic: Performance Evaluation and Comparison

Abstract

Researchers started utilizing and optimizing the state-of-the-art Machine Learning (ML) and Deep Learning (DL) models to benefit Arabic language tools and applications. They employed social media platforms such as Twitter to gather enormous datasets in the Modern Standard Arabic and Arabic dialects, then used the collected datasets to train their models. This noticeable development in the field needs a detailed comparison study to review the work done and highlight the openings for future contributions and improvements. Based on the conducted review, there exists a gap in the time-complexity evaluation of the used ML algorithms in the field of Arabic Sentiment Analysis. Thus, this study presents an experimental approach for determining the time complexity of seven popular ML algorithms in classifying positive and negative Arabic sentences. The results show that the Multi-Layer Perceptron (MLP) and the Support Vector Machine (SVM) have the highest complexity, whereas the Logistic Regression (LR) has the lowest.