Evaluating sentiment analysis for Arabic Tweets using machine learning and deep learning

Abstract

Sentiment analysis is concerned with determining whether a certain material contains online information which expresses positive or negative sentiments. The tools for performing this analysis should be able to identify and assess thoughts and feelings with a reasonable degree of accuracy on feelings that are made openly available by people. It is expected that sentiment analysis would be performed for social media. That is why this paper investigates online social media, as sentiment analysis has become an important subject, and it is one of the approaches employed in the field of natural language processing. Sentiment analysis was applied for an Arabic Twitter dataset in order to identify the feelings expressed by the textual tweets and determine whether they were positive, negative, or neutral. Bigrams and unigrams were used when employing the multinomial Naïve Bayes, Gaussian Naïve Bayes, Logistic Regression, and Support Vector Machines (SVM) machine learning algorithms. The Logistic Regression algorithm achieved the highest accuracy, that is with 63.40%. The Long Short-Term Memory (LSTM) neural network was used for the deep learning-based classification, and it reached an accuracy rate of 70%, a figure which proved to be higher than the results shown in the related works.

Keywords

• sentiment analysis (sa)
• machine learning (ml)
• deep learning