Developing a Browser Extension for the Automated Detection of Deceptive Patterns in Cookie Banners
Juris Hannah Adorna, Aurel Jared Dantis, Rommel P. Feria, Ligaya Leah Figueroa, Rowena Solamo · Atlantis Highlights in Computer Sciences/Atlantis highlights in computer sciences · 2024
Interacting with web-based interfaces is often done with a particular objective in mind.However, deceptive patterns could interfere with these interactions by taking advantage of cognitive biases to either distract users from this objective or mislead them into non-ideal outcomes.These are found in cookie banners that nudge users to allow the tracking of their browsing patterns, infringing upon the user's right to informed consent regarding matters of their privacy.This paper discusses the implementation of a browser extension, Ariadne, that counteracts this by flagging deceptive patterns in cookie banners, in effect safeguarding the user's right to informed consent in data collection.The current implementation is divided into four units: a Naïve-Bayes model determining language clarity (Calliope), a convolutional neural network (CNN) based on VGG-19 determining option weight (Janus), an application programming interface (API) handling the classification and user reports (Dionysus), and the browser extension itself that allow these units to reach the user.The classifiers Calliope and Janus achieved respective accuracies of 85.00% and 100.00% upon unit validation and 80.00% and 66.67% upon unit testing.Integration testing resulted in an overall average accuracy of 68.70% based on the behavior of the browser extension given selected websites as recorded by thirty (30) observers.Acceptance testing was done through an alpha testing questionnaire yielding positive ratings from thirty (30) testers.This project intends to contribute to the larger body of knowledge surrounding the automated detection of deceptive patterns by implementing previous frameworks thereof and setting the groundwork for the creation of a system that can act as a toolbox of methods for the automated detection of deceptive patterns and corresponding methods for intervention.