On Unsupervised and Supervised Discretisation in Mining Stylometric Features

Abstract

Writing styles can be described by stylometric features. They are quantitative in nature, often continuous, which either limits techniques used in mining to those capable of calculations on this form, or adds discretisation to initial pre-processing of data. The paper describes research on unsupervised and supervised discretisation applied in the stylometric domain for the task of authorship attribution. The recognition of authorship is executed as classification performed by chosen inducers capable of operating on both continuous and categorical attributes.