All-or-Nothing Private Record Linkage over Streaming Data

Abstract

The prevalence and increasing need for insights obtained from the collection of sensitive data gives rise to the problem of protecting the privacy of this data. The collection and storage of data can be distributed across locations and organizations, and gaining insights can require combining knowledge from different stores. Private record linkage (PRL) is the problem of finding approximately matching records across different databases while maintaining the privacy of all records involved. The PRL problem in the streaming data model is an emerging problem that tackles PRL in the context of a streaming database, where a service provider performs the matching and learns only the result to gain further insights. To the best of our knowledge our work is the first to address this problem. In this work, we introduce a new cryptographic primitive, the secure approximate equality operator that securely implements all-or-nothing disclosure for approximate matching, which has provable security guarantees in the semi-honest security model. We show that the new operator performs several times faster than a straightforward baseline approach using function-hiding inner product encryption. We also showcase a protocol that implements our new approximate equality operator to perform PRL in the streaming data model with high accuracy and performance.