Anonymization in the Time of Big Data

摘要

In this work we explore how viable is anonymization to prevent disclosure in structured big data. For the sake of concreteness, we focus on fc-anonymity, which is the best-known privacy model based on anonymization. We identify two main challenges to use fc-anonymity in big data. First, confidential attributes can also be quasi-identifier attributes, which increases the number of quasi-identifier attributes and may lead to a large information loss to attain fc-anonymity. Second, in big data there is an unlimited number of data controllers, who may publish independent fc-anonymous releases on overlapping populations of subjects; the fc-anonymity guarantee does not longer hold if an observer pools such independent releases. We propose solutions to deal with the above two challenges. Our conclusion is that, with the proposed adjustments, fc-anonymity is still useful in a context of big data.