A Linguistic System for Predicting Sentiment in Arabic Tweets

The term sentiment analysis is considered very important in our current era, especially with widespread of social media, as it helps understand people's feelings, behavior and opinions about a specific behavior or entity, individuals, organizations and any related topic. Recently, with the development of machine learning, there have been many studies concerned with analyzing feelings. Still, most of these researches are concerned with the English language more than other languages. This paper proposes a model for working with Standard Arabic and some other Arabic dialects such as Levantine, Egyptian, and Gulf. Working with the Arabic language poses several challenges due to the complex structure of the language, the large number of dialects used and the lack of associated resources. The data collected was divided into positive, negative, and neutral. Several algorithms were used to predict sentiment in Arabic texts such as Naive Bayes classifiers (NB), Support Vector Machine (SVM), Random Forest Classifier, and BERT model (Bidirectional Encoder Representations from Transformers). The results obtained are very encouraging, especially with the Bert model (Bidirectional Encoder Representations from Transformers) that gave very accurate results during the test, reaching more than 83%.