JGAAP: A System for Comparative Evaluation of Authorship Attribution

"Nontraditional" authorship attribution, the process of inferring authors or authorial traits via statistical analysis of text, has seen something of a resurgence in recent years. Many methods have been proposed, and most work (after a fashion). However, the sheer number of methods proposed and the sheer number of test corpora developed makes it diffcult to identify any clear "best practices" or particularly accurate techniques. The techniques that appear to score best, such as Support Vector Machines, can be diffcult to understand and hard for non-technical specialists to apply. We will describe and demonstrate a program to address some of these issues. JGAAP (Java Graphical Authorship Attribution Program, available for download from http://www.jgaap.com) is a freely available Java program to perform authorship attribution. Using this program, we describe some advances we are making toward the development of such best practices.