The semantics of persuasion: a case study using phishing emails
University of New Brunswick
As of 2021, phishing emails continue to be the primary means by which network breaches are facilitated. Notwithstanding the development of many tools to detect and block incoming phishing emails, many users continue to be plagued by them on a daily basis. In addition, the nature of phishing emails is changing as the incidence of more personalized forms, such as spear phishing and whaling, prove their effectiveness. These newer forms of phishing are harder to detect using traditional methods and emphasize the need for approaches which seek to enable detection based on persuasion based language features unique to phishing emails. To that end, this thesis draws insights from the phishing process, the applicable behavioural psychology research on persuasion, as well as linguistics, to inform an understanding of how phishing emails persuade. It then proposes a methodology for feature engineering of persuasion language related features for the phishing email domain, based on these insights. A proof of concept model is developed using persuasion based language features, and then implemented and tested using several machine learning algorithms. The performance of this model is as good, if not slightly better, than other more complex and labour intensive efforts which sought to capture semantic meaning using fewer detection features. The thesis concludes with a discussion of potential future work.