One of the most interesting technical presentations I attended in 2012 was the talk on “adversarial stylometry” given by a Drexel College research team at the 28C3 conference in Berlin. “Stylometry” is the practice of trying to ascribe authorship to an anonymous text by analyzing its writing style; “adversarial stylometry” is the practice of resisting stylometric de-anonymization by using software to remove distinctive characteristics and voice from a text.
Stanford’s Arvind Narayanan describes a paper he co-authored on stylometry that has been accepted for the IEEE Symposium on Security and Privacy 2012. In On the Feasibility of Internet-Scale Author Identification (PDF) Narayanan and co-authors show that they can use stylometry to improve the reliability of de-anonymizing blog posts drawn from a large and diverse data-set, using a method that scales well. However, the experimental set was not “adversarial” — that is, the authors took no countermeasures to disguise their authorship. It would be interesting to see how the approach described in the paper performs against texts that are deliberately anonymized, with and without computer assistance. The summary cites another paper by someone who found that even unaided efforts to disguise one’s style makes stylometric analysis much less effective.
We made several innovations that allowed us to achieve the accuracy levels that we did. First, contrary to some previous authors who hypothesized that only relatively straightforward “lazy” classifiers work for this type of problem, we were able to avoid various pitfalls and use more high-powered machinery. Second, we developed new techniques for confidence estimation, including a measure very similar to “eccentricity” used in the Netflix paper. Third, we developed techniques to improve the performance (speed) of our classifiers, detailed in the paper. This is a research contribution by itself, but it also enabled us to rapidly iterate the development of our algorithms and optimize them.
In an earlier article, I noted that we don’t yet have as rigorous an understanding of deanonymization algorithms as we would like. I see this paper as a significant step in that direction. In my series on fingerprinting, I pointed out that in numerous domains, researchers have considered classification/deanonymization problems with tens of classes, with implications for forensics and security-enhancing applications, but that to explore the privacy-infringing/surveillance applications the methods need to be tweaked to be able to deal with a much larger number of classes. Our work shows how to do that, and we believe that insights from our paper will be generally applicable to numerous problems in the privacy space.
Is Writing Style Sufficient to Deanonymize Material Posted Online?
(via Hack the Planet)