Effects of Language Processing in Turkish Authorship Attribution

Abstract

In this study we present a text categorization approach for stylometric analysis of Turkish documents. To this end we extend traditional features and take advantage of linguistic processing. We experiment with a combination of bag of stems and additional features such as function words, part of speech tags and morpho-syntactic tags in datasets having varying number of authors. Based on the characteristics of Turkish (agglutinative language) we expected morpho-syntactic tags to perform better. However, neither part of speech tags nor morpho syntactic tags has showed a significant gain in our settings. Our findings suggest that the main performance is dominated with bag of stem features and the best performance is achieved with combination of bag of stems and function words.