Perils of Location Tracking? Personalized and Interpretable Privacy Preservation in Consumer Mobile Trajectories

Abstract

Consumer location tracking is becoming omnipresent on mobile devices, producing massive volumes of behavior-rich location data and an elaborate multi-billion dollar ecosystem consisting of consumers providing data, third-party data collectors aggregating data, and advertisers using the data for targeting. While such entities can benefit from accessing location data, potential privacy risks do exist. This calls for the data collector to develop an effective privacy-preserving framework before sharing data with advertisers to balance the privacy risks to consumers and data utility to advertisers. We hence propose a personalized framework for the data collector to quantify the consumer-level privacy risks and perform consumer-level data obfuscation to accomplish the privacy-utility trade-off. The framework is also flexible and generalizable to accommodate a variety of risks and utilities. We illustrate the power of the framework with two specific privacy risks (inference of a sensitive attribute such as home location, and re-identification of a consumer from a subset of locations known apriori) and two utilities (location prediction and activity-timing prediction) with machine learning methods. Validating the framework on one million consumer trajectories, we demonstrate the potential privacy risks in the absence of data obfuscation. For instance, an individual's home location can be predicted with 84% success, or within an average radius of 2.5 miles; a consumer (and hence his/her entire trajectory) can be fully identified with 49% success by knowing merely two randomly sampled locations visited by the consumer. Outperforming multiple baselines from the latest literature, the proposed framework significantly reduces each consumer's privacy risk (e.g., by 15% in home inference) while preserving an advertiser's utility. As novel and powerful mobile location data become increasingly leveraged by business leaders, this research offers a personalized and generalizable framework that balances privacy risks and data utilities to sustain a healthier ecosystem based on location data.