Several works of Japanese novelist Kikuchi Kan are thought to be ghostwritten. Serialized novel Junange is one of them. According to the testimony of Kawabata Yasunari, Junange was written by Kikuchi Kan’s pupil, Yokomitsu Riichi. However, this claim remains unsubstantiated due to lack of evidence. In this paper, we verify Junange’s authorship using stylometric methods. We extracted three stylistic features from Junange and 64 novels written by Kikuchi and Yokomitsu. The three stylistic features, which have been reported effective in authorship attribution, are Usage of Comma, Part-of-speech bigram, and Phrase Pattern.
After converting these stylistic features matrices to the relative frequency ones by each work, we used hierarchical clustering analysis and principal component analysis as unsupervised methodologies, and integrated a classification algorithm comprising 7 strong classifiers, including support vector machine, random forest and XGBoost, as the supervised one, to define the authorship of Junange.
According to the results of the analyses mentioned above, we concluded that real author of Junange was Kikuchi Kan.
View full abstract