Automating the Detection and Analyzing the Manipulative Effects of Dark Patterns
Evolving web user interface designs seeded the need to comprehend human perception of user interfaces and to create user behavior models that allow to tailor user interfaces and user experiences and achieve the intended aesthetic and interaction qualities. Initially, these designs were built with simplicity and understated graphics, but with the increased complexity of web applications and increased commercial interest in the web platform, designs gradually adopted persuasive approaches to influence human behavior and decision making, particularly in the domain of E-commerce. Ongoing HCI research is uncovering various deceptive designs, referred to as Dark Patterns. These designs utilize psychological factors favoring business goals over user intentions deliberately. Users are often not aware of these techniques used to manipulate their behavior to maximize conversion rates and profit. Detecting Dark Patterns manually is time-consuming and requires a good understanding of human perception.
This thesis aims at creating an automated approach for the detection of Dark Patterns in E-commerce websites. To achieve this, commonalities and evolution of Dark Patterns in E-commerce platforms need to be investigated and suitable detection requirements derived. Addressing the heterogeneity of web applications, the thesis seeks to answer the question of whether automated detection is possible based on visual aspects. The solution, therefore, needs to accept screenshots of web user interfaces as input and produce detection results along with confidence scores and further details on the identified patterns. To support end users, UI designers, and HCI researchers, the solution architecture needs to incorporate appropriate interfaces. This thesis comprises all necessary steps to create a visual detector for Dark Patterns in E-commerce websites, including data gathering, labeling, training, integration of the models, and evaluation.
The objective of this thesis is to find an approach or a combination of approaches to solve the above-mentioned problem of detection of Dark Patterns in E-commerce websites in the context of Machine learning, Computer Vision, and Natural Language Processing. This particularly includes the state of the art of Dark Patterns and their detection, as well as relevant methods of these three contextual fields. The demonstration of feasibility through an implementation prototype of the concept is a part of this thesis together with a suitable evaluation of the detection performance.