2021 Theses Bachelor's

Predictive Privacy: Modeling Privacy Harms

Radway, Sarah

This work examines how we can account for harm arising from innovation in machine learning and database matching, and will propose a prefatory method for the quantification of harm. The work’s primary contribution is to propose a method of modeling the harm of a dataset, analyzing the risk of deanonymization given these new factors. We carry out two experiments in order to demonstrate the harm accompanying modern dataset applications -- the first exploring machine learning's applications to WhatsApp social network analysis, the second exploring the potential for large-scale database matching between University directories and TinderU profiles. We apply our model to these examples, to demonstrate its the efficacy in a real world setting. We show that this model can be applied as a general framework, to guide both legal regulation of data release and implementation of current anonymization methods.