Learning Personalized Privacy Preference from Public Data

Learning consumers’ personalized privacy preferences is crucial for firms and policymakers to establish trust and compliance and guide effective policymaking. Existing approaches rely mostly on private information such as proprietary user behavior data and individual-level demographic and socio-economic factors, or require explicit user input, which can be invasive and burdensome, potentially leading to user dissatisfaction. Nowadays, individuals generate and share vast amounts of information about themselves in the public domain, which can provide a valuable multifaceted view of their behaviors, attitudes, and preferences. This information thus has the potential to provide valuable insights into individuals’ privacy preferences. In this study, we propose a novel framework to predict personalized privacy preference by leveraging a ubiquitous source of public data—social media posts. Deeply rooted in psychological and privacy theories, we use deep learning model and natural language processing algorithms to learn theory-driven psychosocial traits such as lifestyle, risk preference, personality, privacy-related economic preferences, linguistic styles, and more from social media posts. Interestingly, we find that psychosocial traits from public data provide greater predictive power than private information. Furthermore, we conduct multiple interpretability analyses to understand what drives the model’s performance. Finally, we demonstrate the practical value of our model and show that our framework can assist platforms and policymakers in forecasting the consequences of privacy policies. Overall, our framework provides managerial implications for enhancing consumer privacy control and trust, optimizing platform data management, and informing policymakers about better data privacy regulations.

History: Ravi Bapna, Senior Editor; Heng Xu, Associate Editor.

Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2023.0318.