Turkish Character Usage in Text Classification

Abstract This study is prepared to examine the effects of Turkish character usage on text data by using multiple classifiers. Regression Classifiers, SVM, NB-Classifiers, and ANN are frequently used in supervised learning methods, especially in classification problems. Regression classifiers generally come in two types: as Linear and Logistic. There are also more than one type of Naive Bayes classifier. In our study, after mentioning the properties of Linear Regression and Logistic Regression classifiers in general terms, why Logistic Regression is much more suitable for this study is explained. Then, with the usage of "Logistic Regression", "LinearSVC", "MultinomialNB", "ComplementNB", "BernoulliNB" and "Perceptron" classifiers, the analyzing part starts. Our datasets consist of abstracts-parts from 64 Turkish articles, which have 4 different classes as Physical Sciences, Social Sciences, Educational Sciences, and Economics Administrative Sciences. The data files are all in CSV file format, however, two different data files were prepared. One with original Turkish characters, and the other with its English equivalent formation targeting the Turkish characters "Ç, ç, Ö, ö, Ü, ü, Ş, ş, İ, ı, ğ". In its English-like equivalent file, these were replaced with "C, c, O, o, U, u, S, s, I, i, g" respectively.