The Importance of preprocessing in Turkish Text Classification

Abstract

In this study, the effects of the application of stop words filtering and stemming methods on the classification of Turkish Texts. The documents in a corpus that consists of summaries of conference and journal articles classified by Naive Bayes, Support Vector Machines and Random Forests methods and their performers have been compaired. All the models that have employed preprocessing with stemming and stop words elimination have yielded between 2.26% and 4.94% improvement in performance to the models that have not employed such preprocessing.