Adversarial Authorship, AuthorWebs, and Entropy-Based Evolutionary Clustering
Siobahn Day, J. Brown, Zachery Thomas, India Gregory, Lowell Bass, Gerry Vernon Dozier · 2016
Over the last few years, we have seen an increase in the number of Anonymous Social Networks (ASNs). What many internet users may not know is that their writing style can be tracked across the internet and even through an ASN. The good news is that by using a technique referred to as Adversarial Stylometry one can effectively imitate the writing style of another or even obfuscate their own writing style in an effort to conceal their true writing style - for the short term. The bad news is that recent research has shown that Adversarial Stylometry is not effective in concealing ones writing style over the long term. In this paper, we introduce a number of underlying concepts that will allow users to conceal their writing style over the long term. One such concept we refer to as Adversarial Authorship. In Adversarial Authorship, authors are provided an AuthorWeb which allows them to see graphically how their writing style compares with others in the AuthorWeb. The AuthorWeb presented in this paper uses Entropy-Based Evolutionary Clustering (EBEC) in an effort to cluster writing styles. Our results show that EBEC outperforms a number of other machine learning techniques for author recognition. Users of an AuthorWeb can then write to user-specified clusters in an effort to conceal their writing style.