Visualization of Textual Data: a Complement to Authorship Attribution

Abstract

In textual data analysis, authorship attribution is precisely a leading case of statistical decision. While analyzing a large corpus of 50 French novels of the 20th century, we investigate the frontiers between descriptive (or unsupervised) methods, and confirmatory (or supervised) methods. It will be shown that Additive Trees applied to the coordinates of a preliminary Correspondence Analysis (CA) can provide both a description and a decision. Our results aim at showing the complementarity between exploratory techniques and I.A. in that field.