Political Arabic Articles Classification Based on Machine Learning and Hybrid Vector

Abstract

Recently, there was substantial growth in the opinion data and the number of weblogs in the world wide web (WWW). The capability for automatically determining an article’s political orientation might be of high importance in various fields ranging from academia to security. Yet, sentiment classification related to the weblog posts (especially the political ones), has been more complex in comparison to sentiment classification related to the conventional text. In the presented study, supervised machine learning along with feature extraction methods Term Frequency (TF) and five grams (unigram, bigram, trigram, 4-gram, and 5-gram) were combined to generate a hybrid vector that applied for the process of classification. Besides, for investigation purposes, Support Vector Machine (SVM), Naive Bayes (NB), KNearest Neighbor (KNN), and Decision Tree (DT) for the supervised machine learning were used. After conducting the tests, the results indicated that the NB with unigram provided results with accuracy (93.548%). Thus, the NB is extremely acceptable in the presented model.