Text Analysis in Adversarial Settings: Does Deception Leave a Stylistic Trace?

Textual deception constitutes a major problem for online security. Many have argued that deceptiveness leaves traces in writing style, which be detected using text classification techniques. By conducting an literature review of existing empirical work, we demonstrate that certain linguistic features have been indicative of deception in certain, they fail to generalize across divergent semantic domains. We suggest deceptiveness as such leaves no content-invariant stylistic trace, and similarity measures provide superior means of classifying texts as deceptive. Additionally, we discuss forms of deception beyond content, focusing on hiding author identity by writing style. Surveying the literature on both author identification and techniques, we conclude that current style transformation methods to achieve reliable obfuscation while simultaneously ensuring semantic to the original text. We propose that future work in style should pay particular attention to disallowing semantically changes.