גהאד אלצאנע

This paper describes the impact of dataset characteristics on the results of Arabic document classification algorithms using TF-IDF representations. The experiments compared different stemmers, different categories and different training set sizes,
and found that different dataset characteristics produced widely differing results, in one case attaining a remarkable 99% recall (accuracy). The use of a standard dataset would eliminate this variability and enable researchers to gain comparable knowledge from the published results