Authorship attribution for forensic investigation with thousands of authors

Abstract

With the popularity of computer and Internet, a growing number of criminals have been using the Internet to distribute a wide range of illegal materials and false information globally in an anonymous manner, making criminal identity tracing difficult in the cybercrime investigation process. Consequently, automatic authorship attribution of online messages becomes increasingly crucial for forensic investigation. Although researchers have got many achievements, the accuracies of authorship attribution with tens or thousands of candidate are still relatively poor which is generally among 20%~40%, and cannot be used as evidence in forensic investigation. Instead of asserting that a given text was written by a given user, this paper proposes a novel authorship attribution model combining both profile-based and instance-based approaches to reduce the size of the candidate authors to a small number and narrow the scope of investigation with a high level of accuracy. To evaluate the effectiveness of our model, we conduct extensive experiments on a blog corpus with thousands of candidate authors. The experimental results show that our algorithm can successfully output a small number of candidate authors with high accuracy.