UCLA researchers in the Department of Computer Science have developed a novel way in which to secretly search for and collect relevant information from a streaming database. The invention has application to intelligence gathering and data mining.
The ability to secretly search for and return important parts of a streaming database has many applications. For instance, one of the central tasks of the intelligence community is to secretly collect and analyze real-time information that is relevant to national security. Online marketers would appreciate a tool that would allow them to actively discern consumer purchasing behavior without influencing it. Also, the ability to search through medical transactions in a way that would still protect the confidential portions of those records would be of great benefit.
Tools do exist that allow for the secret search of data sources. However, in order to remain secret, these tools must first acquire and store large amounts of information and later analyze that information in a secure environment. Such an approach is expensive and resource intensive as it requires the constant monitoring and/or storage of large amounts of data, much of which will ultimately be deemed irrelevant. A more efficient solution would be to search and “flag” data as it is generated at its source, i.e. a document on a server or a packet on a router. In this way, only relevant or flagged data would be transmitted for further analysis, resulting in a considerable reduction in data monitoring and storage cost. The invention disclosed here offers a means to secretly search, flag, and return important parts of streaming data as it is generated at its source, even if that source is unclassified.
A secret, localized search technique allows for efficient, private keyword search of distributed, streaming data, typical of online (internet) environments, emails, user requests, etc. The localized search process is hidden from the computer that executes the program and it is provably impossible to mount reverse-engineering attempts to reveal the secret search.
The innovation has been fully conceptualized and its correctness and security have been proven.
|United States Of America||Issued Patent||8,291,237||10/16/2012||2005-487|
Private Information Retrieval, Single Database Private Keyword Search, Algorithms for Streaming Data, Privacy-preserving Data Mining