Sentiment extraction from arabic social media: A case study on modern standard arabic and Jordanian tweets

Over the recent past, social media has been embraced as a source of sentiment and emotion data for opinion and behavior analysis. This research focuses on the sentiment classification of Arabic tweets and the Naive Bayes, Support Vector Machine (SVM), Random Forest, Logistic Regression, and other models. The goal is to classify the tweets into the positive and negative sentiment. The classifiers performances were measured in terms of accuracy, precision, recall, and F1-Score. The classifiers Logistic Regression and Random Forest were more accurate than others with 84% and 79.5% respectively whereas K-Nearest Neighbors were lagging behind with only 58% accuracy. The findings show the viability of traditional supervised machine learning models for the sentiment analysis of Arabic text, but with the possibility of enhancement with AraBERT and other deep learning models. Thus, the results of this work can help to advance the methods for building the sentiment analysis for Arabic, which is a language with relatively scarce resources for training the models.