Privacy-preserved distinct content collection in human-assisted ubiquitous computing systems

Abstract

Abstract Human beings with smart devices have dramatically facilitated the knowledge acquisition in ubiquitous computing systems. However, the collected contents also severely threat the security and privacy for individuals, as these contents achieve a fine-grained coverage of one’s behaviors and movements. Local differential privacy, as one of the widely accepted properties, enables individuals to contribute to data collection while remaining indistinguishable against adversaries. However, existing solutions and mechanisms fail to capture the contents with relatively low frequency, e.g., the temperature in local regions visited by very few people. Therefore, this work proposes a novel framework for distinct content estimation under local differential privacy. We propose a hash function and random response based framework for the estimation. The framework allows individuals in the ubiquitous computing system to participate with flexible bandwidth, and properly assigns the given bandwidth to improve estimated results. We prove the optimization problem of assignment in our framework to be NP-complete, and provides an effective heuristic algorithm. Our framework is also proved to guarantee an unbiased estimation for distinct contents, and achieves local differential privacy for users. Meanwhile, we also propose a randomized algorithm designed for more general cases, where the preposed knowledge for the first algorithm is unavailable. Finally, we evaluate both algorithms on real world datasets, and the results reveal that our algorithms outperform existing methods.