Analyzing readability level of educational content in Turkish language

In today´s environment, in which digital data is continuously increasing, it is of prime importance for students to find data appropriate for their readability level. In this study, our aim is to classify educational data in Turkish based on their readability level. Three readability formulas and new syllable and word level features are used in this study. K12 level Turkish Language course textbooks published by Turkish Ministry of National Education are used as training data. Classifier models are created with Naïve Bayes, Decision Tree, Random Forest and Multilayer Perceptron classification algorithms, from these books. Our test data includes educational web pages obtained from 14 different web sites. As a result of the study, the readability formulas with the suggested word level features achieved more successful readability level detection than the readability formulas without them.