A Comprehensive Analysis for Dark Pattern Detection Using Structural, Visual and Textual Information

Abstract

Dark patterns are deceitful design strategies that control and influence user behavior and have become widespread in digital interfaces across various domains. In this research paper, dark patterns have been classified into broad types: structural dark patterns and UI-based (textual and visual) dark patterns. We present three parallel approaches for detecting the structural and vision based dark patterns effectively. DOM inspector is used for detecting and classifying structural-based dark patterns. For detecting visual dark patterns, YOLOv5 is applied. And finally, for detecting and classifying text-based dark patterns, an EasyOCR and DistilBERT-based approach is presented. We study various websites consisting of different subjects of interest, including e-commerce, news, sports, and business, to analyze the prevalence and types of dark patterns employed. The results and outcomes help us shine a light on the prevalent nature of dark patterns in digital media and provide insights into detecting and reducing their impact on user experience, which will in turn help build trust.