In this study, we have created a new Arabic dataset annotated according to Ekman’s basic emotions (Anger, Disgust, Fear, Happiness, Sadness and Surprise). This dataset is composed from Facebook posts written in the Iraqi dialect. We evaluated the quality of this dataset using four external judges which resulted in an average inter-annotation agreement of 0.751. Then we explored six different supervised machine learning methods to test the new dataset. We used Weka standard classifiers ZeroR, J48, Naïve Bayes, Multinomial Naïve Bayes for Text, and SMO. We also used a further compression-based classifier called PPM not included in Weka. Our study reveals that the PPM classifier significantly outperforms other classifiers such as SVM and Naïve Bayes achieving the highest results in terms of accuracy, precision, recall, and F-measure.
