Detecting dark patterns in shopping websites – a multi-faceted approach using Bidirectional Encoder Representations from Transformers (BERT)

Dark patterns refer to certain elements of the user interface and user experience that are designed to deceive, manipulate, confuse, and pressure users of a particular platform or website into making decisions they wouldn't have made knowingly. Many companies have begun implementing dark patterns on their websites, employing carefully crafted language and design elements to manipulate their users. Numerous studies have examined this subject and developed a classification system for these patterns. Additionally, governments
worldwide have taken actions to restrict the use of these practices. This proposed work seeks to establish a fundamental framework for developing a browser extension, the purpose of which is to extract text from a specific shopping website, employ Bidirectional Encoder Representations from Transformers (BERT), an open-source natural language processing model, to identify and expose dark patterns to users who may be unaware of them. This tool's development has the potential to create a more equitable environment
and enable individuals to enhance their knowledge in this area. The proposed work explores the issues and challenges associated with detecting dark patterns, as well as the strategies employed by companies to make detection more challenging by carefully modifying the design of their websites and applications. Moreover, the proposed work aims to enhance the accuracy for the detection of dark patterns using a natural language processing (NLP) model i.e, BERT which results in accuracy 97% compared to classical models such as Random Forest and SVM having accuracy of 95.4% and 95.8% respectively. It seeks to facilitate future research and improvements to ensure the tool remains up-todate with the constantly changing tactics