Describir: Comparative evaluation of text classification techniques using a large diverse Arabic dataset.