Analysis of Intertextual Distances Using Multidimensional Scaling in the Context of Authorship Attribution

Abstract

Four distance functions were evaluated in order to determine which one better represents two types of style markers (named as static and dynamic) commonly used for authorship attribution tasks. Intertextual distances were analysed from different authors and evaluated if the closest text to another was written by the same author. Classic multidimensional scaling was used to visualize intertextual distances because we consider that this is a method that allows the judges to better understand and visualize the results.

The outcome of this paper is that selecting different distance functions considering the type of style marker improves the clustering of texts from the same author. We concluded that while for static features Canberra distance is recomendable, the dynamic features must depend on the style of each author.