Abstract
The Internet has a huge amount of information when it comes to analysis, much of which is valuable and significant. Arabic Sentiment Analysis (SA) is a method responsible for analyzing people’s thoughts, feelings, and responses to a variety of products and services on social networking and commercial sites. Several researchers utilize sentiment analysis to determine the opinions of customers in various areas, including e-marketing, business, and other fields. Deep learning (DL) is a useful technology for developing sentiment analysis models to improve e-marketing operations. There are a few studies targeting Arabic sentiment analysis (ASA) in e-marketing using deep learning algorithms. Due to a number of difficulties in the Arabic language, such as the language’s morphological features, the diversity of dialects, and the absence of suitable corpora, sentiment analysis on Arabic material is restricted. In this paper, we will compare several Arabic sentiment analysis models. Also, we discuss the deep learning algorithms that are employed in Arabic sentiment analysis. The domain of the collected papers is Arabic sentiment analysis in e-marketing using deep learning. Our first contribution is to introduce and present deep learning models that are used in ASA. Secondly, investigate and study Arabic datasets utilized for Arabic sentence analysis. We create and develop a new Arabic dataset for Saudi Arabian communication companies, namely Sara-Dataset, to increase the quality and quantity of their services. Third, each collected study is assessed in terms of its methodology, contributions, deep learning techniques, performance, Arabic datasets in emarketing, and potential improvements in developing Arabic sentiment analysis models. Fourth, we analyzed several papers’ performance in terms of accuracy, F-measure, recall, pre-procession, and area under the curve (AUC). Also, our comparative analysis includes feature selection (e.g., domain-specific selection) methods that are used in Arabic sentiment analysis. Fifth, we also discuss how to improve Arabic sentiment analysis using preprocessing techniques (e.g., word embedding). Finally, we provide a design model for analyzing Arabic sentiment about communications services provided by Saudi Arabian enterprises.

Key words: Deep Learning; Comparative; Arabic Sentiment Analysis; E−marketing; Accuracy; Dataset; Feature Selection; Pre−processing CNN; LSTM