Preserving Access Privacy Over Large Databases

Anonymity systems preserve the identities of users as they access Internet data sources. The security of many such systems, such as Tor, relies on a model where the adversary does not have a global view or control of the network. A different problem is that of preserving access privacy for users accessing a large database over the Internet in a model where the adversary has full control of the database. Private information retrieval (PIR) has been introduced to prevent such a powerful adversary from being able to learn any information about the user’s query, such as which particular one of the many data items under the adversary’s control has been retrieved. However, the state-of-the-art PIR schemes have a high computational overhead that makes them expensive for querying large databases. In this paper, we develop an access privacy technique and system for querying large databases. Our technique explores privacy-preserving constraint-based query transformations, offline data classification, and privacy-preserving queries to index structures much smaller than the databases. We draw inspiration from techniques in the traditional information retrieval domain. Our approach allows users to query a large database by statically specifying or dynamically defining database portions with inherently high diversity, thereby minimizing information leakage about the data items of interest. Unlike naive approaches where database portions are defined without considering privacy-maximizing subsets, we propose a technique for improving the privacy derivable from portions of the database used for answering user queries. In addition, our approach requires minimal user intervention and allows users to specify descriptions of their privacy preferences and cost tolerances along with their input queries to derive transformed queries capable of satisfying the input constrains when executed. We evaluated the system using patent data made available by the United States Patent and Trademark Office through Google Patent; however, the approach has a much wider application and the system developed can be adapted and deployed for use with many user-centric privacypreserving systems, thereby making access privacy obtainable for today’s Internet users.