Detecting Deceptive Dark-Pattern Web Advertisements for Blind Screen-Reader Users

Area of Science:

• Human-Computer Interaction
• Web Accessibility Research
• Digital Advertising

Background:

• Web advertisements are prevalent but primarily designed for visual interaction.
• The impact of advertisements on blind users relying on screen readers is under-researched.
• Existing studies on non-visual web interaction overlook the specific challenges posed by ad content.

Purpose of the Study:

• To investigate how contextually deceptive advertisements affect blind users' web browsing experience.
• To develop and evaluate an automated method for identifying contextually deceptive ads.
• To enhance web accessibility for visually impaired users by addressing ad-related barriers.

Main Methods:

• Conducted an interview study with 18 blind participants to understand their experiences with web ads.
• Developed a detection model using a multi-modal approach combining handcrafted and automatically extracted features.
• Evaluated the model's performance on a test dataset and real-world websites.

Main Results:

• Blind users reported being frequently deceived by ads that blend seamlessly with website content.
• The developed algorithm achieved high detection accuracy, with F1 scores of 0.86 on a test dataset and 0.88 on live websites.
• Identified limitations of ad blockers, particularly their ineffectiveness against internally injected ads and website access restrictions.

Conclusions:

• Contextually deceptive advertisements pose a significant challenge to blind web users.
• The proposed detection algorithm offers a viable solution for identifying and potentially mitigating deceptive ads.
• Further research and development are needed to improve the accessibility of web advertising for visually impaired individuals.