Supervised classifiers and keyword extraction methods for text classification in Arabic

Text classification categorizes a text document into one or more pre-defined classes based on its content. It is commonly used to organize online documents into categories for easier information extraction. Most studies in this area have used supervised learning techniques. The features in supervised text classification are the words present in the training documents. The more features there are, the higher the dimensionality of the dataset becomes. However, having too many input features can make the training process difficult and time-consuming. To tackle this issue, we can use the most relevant words about the document’s content as features by employing keywords. This helps to reduce the high-dimensional feature space. Text categorization for Arabic needs to include significant research. Therefore, the main objective of this paper is to compare various keyword extraction methods for Arabic news articles in conjunction with machine learning classifiers. Specifically, the study focuses on investigating embedding-based keyword extraction methods. Six classifiers are utilized to conduct a comparative evaluation. These classifiers are Naive Bayes, Decision Trees, Random Forest, Bagging+Naive Bayes, Bagging+Decision Trees, and Voting. The experimental analysis of the study reveals that the Aravec-based keyword extraction method, which is an embedding-based approach, demonstrates the highest predictive performance for classification among all methods. Besides, Naive Bayes and Bagging+Naive Bayes classifiers perform well for most keyword extraction methods. The study further suggests that with embedding-based keyword extraction methods, the predictive performance of classifiers improves as the number of keywords increases. However, for traditional methods, the classifier’s predictive performance improves with fewer keywords.