Classification of writing style by using a morpheme network analysis

Abstract

Writing style is hard to define and classify because it has many different definitions. Some researchers, including Mendenhall, Zipf, Mosteller and Wallace, made attempts to approach the authorship problem in a quantitative way. The problem of writing style consists of three categories: authorship identification, authorship characterization and similarity detection. Here, we focus on similarity detection where we compare and classify unlabeled articles. Text mining techniques, with a slight modification of the method, are used to compare the styles of writing. While the existing method mainly focuses on words, we form a network of authors by analyzing the articles based on morphemes. Articles from the Hankyoreh newspaper were targeted, from which 991 and 951 articles were randomly selected to form two discrete networks. We assumed that human and occupational styles were present in writing style, and through modularity to determine communities, we found three groups in both discrete networks. Our results show the possibility of making an objective definition of style, even though various definitions of style may exist linguistics. © 2018 The Korean Physical Society. All Rights Reserved.