CRI: CRD: Collaborative Research: Community Resources for Authorship Attribution Research

点击翻译按钮获取中文摘要

Homeland security and the criminal and civil justice systems increasingly require reliable and valid methods for automatically identifying the authors of anonymous documents. Further demand for effective author attribution arises from fields as diverse as computer forensics and literary studies. However, despite the clear and growing need for methodological research in this area, there are as yet no standard test suites for authorship attribution, and hence no agreed upon ways to compare research results and validate techniques. This situation, combined with the highly interdisciplinary nature of the field, has led to much redundant and sometimes unsound research.The goal of this project, therefore, is to develop standards and procedures for annotating corpora for use in authorship attribution research and evaluating the results of such research. To this end, the PIs are developing both a large corpus of emails annotated with information about email authors and recipients (including both identity and sociodemographic information), and a suite of testbed attribution tasks based on this corpus. The corpus will form the basis of a research community evaluation exercise which is being run as part of the project, which serves three purposes ? providing baseline results for future authorship research, integrating the currently diverse research community, and most importantly, ensuring the quality of the resulting authorship corpus annotation standards and evaluation procedures.