Articles

This research introduces DarkDialogs, a system designed to automatically extract consent dialogs from websites and identify 10 types of dark patterns. Evaluated against a hand-labelled dataset, DarkDialogs boasts a 98.7% accuracy in dialog extraction and a 99% accuracy in classifying dark patterns. When deployed on 10,992 websites, the system successfully collected 2,417 consent dialogs and detected 3,744 distinct dark patterns. The study also investigates the correlation between dark pattern prevalence and website popularity, the use of third-party consent management providers, and the number of ID-like cookies.