Generalizing PIR for Practical Private Retrieval of Public Data

Private retrieval of public data is useful when a client wants to query a public data service without revealing the query to the server. Computational Private Information Retrieval (cPIR) achieves complete privacy for clients, but is deemed impractical since it involves expensive computation on all the data on the server. Besides, it is inflexible if the server wants to charge the client based on the service data that is exposed. k-Anonymity, on the other hand, is flexible and cheap for anonymizing the querying process, but is vulnerable to privacy and security threats. We propose a practical and flexible approach for the private retrieval of public data called Bounding-Box PIR (bbPIR). Using bbPIR, a client specifies both privacy requirements and a service charge budget. The server satisfies the client's requirements, and achieves overall good performance in computation and communication. bbPIR generalizes cPIR and k-Anonymity in that the bounding box can include as much as all the data on the server or as little as just k data items. The efficiency of bbPIR compared to cPIR and the effectiveness of bbPIR compared to k-Anonymity are verified in extensive experimental evaluations.