An Information-Theoretic Approach for Privacy Protection in OLAP Systems

We address issues related to the protection of private information in Online Analytical Processing (OLAP) systems, wherea major privacy concern is the adversarial inference of private information from OLAP query answers. Most previous work onprivacy-preserving OLAP focuses on a single aggregate function and/or addresses only exact disclosure, which eliminates fromconsideration an important class of privacy breaches where partial information, but not exact values, of private data is disclosed(i.e., partial disclosure). We address privacy protection against both exact and partial disclosure in OLAP systems with mixedaggregate functions. In particular, we propose an information-theoretic inference control approach that supports a combination ofcommon aggregate functions (e.g., COUNT, SUM, MIN, MAX, MEDIAN) and guarantees the level of privacy disclosure not toexceed thresholds pre-determined by the data owners. We demonstrate that our approach is efficient and can be implemented inexisting OLAP systems with little modification. It also satisfies the simulatable auditing model and leaks no private informationthrough query rejections. Through performance analysis, we show that compared with previous approaches, our approach providesmore effective privacy protection while maintaining a higher level of query-answer availability. Furthermore, our approach can bereadily integrated with perturbation-based mechanisms to improve the performance of both.