Dark patterns are deceptive user interface designs for online services that
make users behave in unintended ways. Dark patterns, such as privacy invasion,
financial loss, and emotional distress, can harm users. These issues have been
the subject of considerable debate in recent years. In this paper, we study
interpretable dark pattern auto-detection, that is, why a particular user
interface is detected as having dark patterns. First, we trained a model using
transformer-based pre-trained language models, BERT, on a text-based dataset
for the automatic detection of dark patterns in e-commerce. Then, we applied
post-hoc explanation techniques, including local interpretable model agnostic
explanation (LIME) and Shapley additive explanations (SHAP), to the trained
model, which revealed which terms influence each prediction as a dark pattern.
In addition, we extracted and analyzed terms that affected the dark patterns.
Our findings may prevent users from being manipulated by dark patterns, and aid
in the construction of more equitable internet services. Our code is available
at https://github.com/yamanalab/why-darkpattern.
Dark patterns are a pressing issue in the design of online services, as they deceive users and can lead to negative outcomes such as privacy invasion, financial loss, and emotional distress. Understanding why certain user interfaces are detected as having dark patterns is crucial for addressing this problem effectively.
In this study, the researchers focused on developing an interpretable dark pattern auto-detection model. They utilized transformer-based pre-trained language models, specifically BERT, to train a model on a text-based dataset related to e-commerce. This training allowed the model to automatically detect dark patterns.
However, simply detecting the presence of dark patterns is not enough to gain insights and address the underlying issues. To provide transparency and interpretability, the researchers applied post-hoc explanation techniques to the trained model. Two prominent techniques used were local interpretable model agnostic explanation (LIME) and Shapley additive explanations (SHAP).
The application of these explanation techniques revealed which specific terms influenced each prediction made by the dark pattern detection model. This information is valuable in identifying the strategies employed by dark patterns and understanding the deceptive techniques they employ. By understanding the specific terms, organizations can better design their user interfaces to prevent manipulation and protect their users.
In addition to identifying influential terms, the researchers also examined and analyzed the terms that affected the detection of dark patterns. This analysis provides valuable insights into the underlying linguistic cues or characteristics associated with dark patterns. Identifying these characteristics can enhance our understanding of how dark patterns are constructed and potentially aid in the creation of more equitable and user-friendly internet services.
The interdisciplinary nature of this research is noteworthy. It combines concepts from various fields such as human-computer interaction, natural language processing, and machine learning. By bringing together expertise from these disciplines, the researchers provide a comprehensive approach to tackle the problem of dark patterns.
The availability of the code used in this research on GitHub (https://github.com/yamanalab/why-darkpattern) is laudable. It allows other researchers and practitioners to replicate the experiments, build upon the work, and contribute to the development of more effective tools for detecting and combating dark patterns.
In conclusion, this study on interpretable dark pattern auto-detection sheds light on the strategies and linguistic cues behind dark patterns in online services. By utilizing advanced language models and explanation techniques, the researchers offer insights that can aid in the prevention of manipulative user interfaces and support the creation of fairer internet services.