A Microsoft research trio has developed an algorithm capable of eliminating user tracking in web search without the overheads of existing technology.
The idea, to be presented next month and titled Bloom Cookies: Web Search Personalisation without User Tracking, uses a new type of flowery cookies that can tightly-encode user profiles to preserve privacy without cutting off online personalisation services.
"The Bloom Cookies design is inspired by our analysis of a large set of web search logs that shows drawbacks of two profile obfuscation techniques, namely profile generalisation and noise injection, today used by many privacy-preserving personalization systems," the trio wrote in an abstract explaining the work.
"We find that profile generalisation significantly hurts personalization and fails to protect users from a server linking user sessions over time.
"Noise injection can address these problems, but only at the cost of a high communication overhead and a noise dictionary generated by a trusted third party."
Bloom Cookies, said Nitesh Mor and John Kubiatowicz of the University of California, and Oriana Riva and Suman Nath of Microsoft Research, would use Bloom filters to score a better tradeoff between privacy, personalisation, and network efficiency.
The filters are space-efficient structures used to confirm elements as parts of a set in a way that eliminates false negatives.
"They provide similar or better personalisation and privacy than noise injection and profile generalisation, but with an order of magnitude lower communication cost and no noise dictionary," they said.
It is unclear just how this approach sits alongside Bing and its tracking features.
Noise injection as a means to enhance search privacy protection has been proposed multiple times, with the first theoretical analysis on its effectiveness done in 2009.
Microsoft's work builds on this line of inquiry and will be published at the NDSS Symposium in San Diego would detail an algorithm capable of tweaking noise within Bloom Cookies according to a user's personalisation preferences. ®