Comparative study of Arabic text classification using feature vectorization methods
Tarik Sabri, Omar El Beggar, Mohamed Kissi
Abstract
Arabic Text Classification (ATC) also known Arabic text categorization is the task of assigning categories to Arabic documents based on their contents. It is mostly used for sentiment analysis, detecting trends in customer feedback, spam detection and topic labeling. This paper presents an empirical study of five classification models using two Arabic datasets cnn_arabic and osac_uft8. These algorithms are Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbor (KNN) and Logistic Regression (LR). Three feature vectorization methods were applied to convert text into numeric vectors word count, Terms Frequency-Inverse Document Frequency (TF-IDF) and word embedding using word2vec. For the applied feature vectorization techniques, the experiment shows that the classifiers SVM and LR score the highest performance followed by RF, KNN and DT. Besides, the experiment shows that feature vectorization methods and dataset size have high impact on the performance of the algorithms RF, KNN and DT, while SVM and LR maintain stable outcomes.