Adversarial Authorship, Sentiment Analysis, and the AuthorWeb Zoo

This paper looks at Doddington's Biometric Zoo from the perspective of AuthorWebs for Adversarial Authorship. We provide a definition of an AuthorWeb and show how animals (authors) in the AuthorWeb Zoo are similar to those found in the Biometric Zoo from a Sentiment Analysis perspective. Additionally, we investigate the use of Sentiment Analysis for Authorship Attribution. When initially experimenting with sentiment analysis features, the accuracy was quite low (18%- 22%). However, applying GEFeS (Genetic & Evolutionary Feature Selection) dramatically increased the accuracy from 22% to 57% while only using 21.6% of the Sentiment Analysis Features.