Toward Privacy Management Automation: A Framework for Assessing Privacy Sensitivity

The increasing prevalence of digital technologies has intensified the tension between the benefits of online engagement and the risks associated with privacy loss. Effective privacy management therefore, requires not only regulatory safeguards but also the ability to evaluate the privacy sensitivity of information prior to its disclosure. This study proposes an automated methodology for assessing the privacy sensitivity of textual content from the user’s perspective. The proposed framework integrates Natural Language Processing (NLP) techniques with Machine Learning (ML) models and incorporates individual-level attributes, including demographic characteristics and privacy attitudes, to generate personalized privacy sensitivity predictions. The methodology is empirically validated using data from the social networking platform Twitter (X), where privacy sensitivity labels were obtained through a large-scale crowdsourcing process. The resulting model demonstrates strong predictive performance, achieving a Mean Absolute Deviation (MAD) of 0.18 on a privacy sensitivity scale ranging from 0 to 2.5 and a Mean Squared Error (MSE) of 0.226 on a scale of 0 to 6.25. By enabling real-time, automated evaluation of privacy sensitivity for free-text disclosures, this research contributes a practical and scalable component for privacy-aware decision-support systems. The proposed framework lays the groundwork for the development of intelligent privacy assistants that support users in managing information sharing in dynamic digital environments.