Arabic Text Classification: An Improved Model using New Relations-Based Features

University of Information Technology and Communication, Baghdad, Iraq

Abstract

As a result of increasing Arabic text documents' warehouses on local PC storage as well as on the Web, various tools are emerged to process this type of documents. Text classification and categorization are the most important tools to classify documents in order to save, sort and retrieve these documents later. Accordingly in this paper, an improved model to classify Arabic text documents is proposed. In this model, relations between concepts in the Arabic WordNet dictionary are utilized to propose five new features. These new features are compared with the state of the art features using three quantitative metrics, three evaluation datasets, and three classification algorithms. In the results, the new Proposed Relation-based Features (PRF) show their superiority on the state of the art features in most cases.