Nowadays, with a vast amount of data being generated in the internet and social media platforms, there is much focus on sentiment analysis as it helps to extract data ultimately to analyze people's opinions. However, much of the research on this topic has been regarding the English language with less attention being paid to other common languages such as Arabic. That is why there is a noticeable gap in research especially when considering the multitude of dialects in the Arabic language. This multitude of dialects is considered to be a double-edged weapon as it offers a rich amount of data for researchers from one hand and it stands as a challenge in front of the annotators and the researchers from another. Accordingly, this article aims to carry out a critical review of the recent studies conducted regarding sentiment analysis and data annotation for Arabic dialects that have been published in the last 7 years. This review offers a taxonomy of data preprocessing and annotation methods. Moreover, it displays the challenges, motivations and recommendations in addition to an in-depth analysis of the current trends in the field of Arabic dialects and sentiment analysis. Accordingly, this literature review postulates new research gaps and future directions to drive more scholars into contributing to Arabic SA research. Finally, the contribution of the current study aims at making more successful multilingual sentiment analysis implementations in addition to providing some understanding of ASA in a plenty of settings.