HAR – 03 – TOR, Stylometry and a happy end

Posted August 19th, 2009 by konrad

Enlighted CCCThere were two other HAR presentations that I like to recommend: At first Roger Dingledine explained the performance issue of the network privacy tool TOR in his talk and the ways of dealing with this in the future. He had a lot of small anecdotes about the project that made the talk really interesting and memorable.

I often thought about the problem of the possibility that people could be identified by their writing style. Especially today as many people generate a huge online available corpus that could be used to train detection algorithm (which are similar to the ones used e.g. in bioinformatics). Accidentally I was stuck (this wasn’t planned but it was my great luck :)) in a talk by Mike Brennan about stylometry that shed light upon this issue. In the first part of his talk Mike presented how sophisticated this field is. There are e.g. unsupervised pairing methods with 95% accuracy even if only with a short text is available. Mike also presented methods to protect against such privacy attack: A very successful method is the imitation of the writing style of another person (e.g. a famous author). He proposed assisting software to increase the success rate of this approach.

If you are interested in one of the many talks just have a look at the stream dumps that are already available. I assume there will be further processed versions in the near future.

Many thanks to everybody who made the HAR such a great event!

Leave a Reply

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>