Emotions extraction from Arabic tweets

ABSTRACT Twitter is one of the most used microblogs in social media communication channels. Emotion detection has recently raised as an important research field. Extracting emotions in Twitter microblogs has many benefits and applications. Such applications include e-commerce, e-marketing, and others. Knowing the perception about relevant products, services, events or personalities, as well as monitoring their online reputation are some of the objectives that companies have marked in short term. Most of studies focus on sentiments analysis as positive and negative but few of them go deeper to analyze and classify the emotions behind tweets, especially in Arabic tweets. Arabic language becomes a hard challenge for emotions classification on twitter and it involves more preprocessing before classification than other languages. This paper presents a model for extracting and classifying emotions in Arabic tweets based on four emotions: sad, joy, disgust, and anger. The experimental results demonstrate the validity of the proposed model, which improves the state of the art in the classification of Arabic tweets using support vector machine (SVM) and Naïve Bayes (NB) that give the best results. SVM outperforms the other used classifiers with 80.6% accuracy, and the NB outperforms the other classifiers with 0.95 ROC area.