Abstract
Social media provides a significant source of public opinions and trends. Recently, the interest in analyzing this publicly available data through sentiment analysis has increased noticeably. The use of deep-learning for sentiment analysis is lately under focus, as it provides a scalable and direct way to analyze text without the need to manually feature-engineer the data. As the work on Arabic sentiment analysis using deep learning is scarce and scattered, this paper presents a systematic review of those studies covering the whole literature, analyzing 19 papers. The review proves a general trend of Arabic sentiment analysis performance improvement with deep learning as opposed to sentiment analysis using machine learning.