Privacy Avatar Paper accepted @ MUM 2025 🎉
Our paper Investigating User Perceptions of Visualising Inferred Sensitive Information as Privacy Avatars by Lena Swienty, Rebecca Panskus, Stefanie Miketta, Beyza Kircili, Sen Cheng was Karola Marky was accepted as full paper at the 24th International Conference on Mobile and Ubiquitous Multimedia (MUM), which will take place in Enna, Italy.
In this paper, we use generative AI to generate a picture of the user (see figure above) based on data shared during online shopping. We consider this a novel method for an intuitive privacy interface to visualize information that algorithms can infer about users based on shared data and website interaction.
🠶 Abstract: Collecting user data is ubiquitous in the modern web, because it enables different kinds of services like personalised recommendations. Specific data shared by the users can be analysed further to infer additional information – e.g., gender, age, appearance, or preferences. Such inferences are challenging to predict for users when deciding on whether or not to share data. In this paper, we explore privacy avatars as a way to visualise collected and inferred data to users by a picture. For this, we conducted semi-structured interviews (N=20) where participants were shown privacy avatars based on products they bought in a fictitious online-shopping scenario and additional privacy avatars with a varied level of detail and information. Our results show that participants were shocked by avatars that accurately depicted parts of their visual appearance and perceived inaccurate avatars as privacy-preserving. However, participants preferred information they considered sensitive (e.g., gender) to be accurate creating a tension between privacy and accuracy of shared or inferred information. We conclude with a discussion on how such privacy avatars could be generated and used in practice to offer intuitive and informative privacy interfaces.
