Automatic categorization of Arabic articles based on their political orientation

The ability to automatically determine the political orientation of an article can be of great benefit in many areas from academia to security. However, this problem has been largely understudied for Arabic texts in the literature. The contribution of this work lies in two aspects. First, collecting and manually labeling a corpus of articles and comments from different political orientations in the Arab world and making different versions of it. Second, studying the performance of various feature reduction methods and various classifiers on these synthesized datasets. The two most popular feature extraction approaches for such a problem were compared, namely the Traditional Text Categorization (TC) approach and the Stylometric Features approach (SF). Although the experimental results show the superiority of the TC approach over the SF approach, the results also indicate that the latter approach can be significantly improved by adding new and more discriminating features. The experimental results also show that the feature selection techniques reduce the accuracies of the considered classifiers under the TC and SF approaches in general. The only exception is the Partition Membership (PM) technique which has an opposite effect. The highest accuracies are obtained when PM feature selection method is used with the Support Vector Machine (SVM) classifier.