Using Frequent Fixed or Variable-Length POS Ngrams or Skip-Grams for Blog Authorship Attribution

Abstract

Authorship attribution is the process of identifying the author of an unknown text from a finite set of known candidates. In recent years, it has become increasingly relevant in social networks, blogs, emails and forums where anonymous posts, bullying, and even threats are sometimes perpetrated. State-of-the-art systems for authorship attribution often combine a wide range of features to achieve high accuracy. Although many features have been proposed, it remains an important challenge to find new features and methods that can characterize each author and that can be used on non formal or short writings like blog content or emails. In this paper, we present a novel method for authorship attribution using frequent fixed or variable-length part-of-speech patterns (ngrams or skip-grams) as features to represent each author's style. This method allows the system to automatically choose its most appropriate features as those sequences being used most frequently. An experimental evaluation on a collection of blog posts shows that the proposed approach is effective at discriminating between blog authors.