Authorship Attribution in Huayan Texts by Machine Learning using N-gram and SVM

Abstract

This paper aims to evaluate the authorship of Buddhist texts in the classical Chinese language, specifically focusing on Fazang’s Huayan texts, using computational statistical methods. I also briefly introduce the basic ideas and processes related to this estimation. Regarding statistical techniques, this paper utilizes n-gram and one-class SVM (Support Vector Machine), a technique of machine learning.BR The authorship attribution (AA) process in stylometry using machine learning can be divided into the following three stages: (1) Building a Corpus → (2) Extracting a Feature Set → (3) Learning and Predicting Authorship.BR From the results of the tests, the following can be estimated.BR (1) This one-class SVM model is somewhat reliable given that training observations and regular tests are distributed in similar regions and abnormal tests are in different regions.BR (2) As the target is located in a different area from the training observations and the regular test area, the target is not likely to be Fazang’s work.BR (3) However, considering the fact that it is closer to training observations and regular tests than an abnormal test by Zhiyan’s writing, it can be inferred that the Huayanjing wenda is closer to Fazang’s writing than Zhiyan’s.BR One thing to keep in mind is that this is only a preliminary test. However, I think that there is enough meaning to show the possibility and direction about AA in Chinese Buddhist literature.