ADDING noise to search-engine records could help keep surfers鈥� identities private. A team from Microsoft Research in Mountain View, California, says the technique is a major step towards 鈥減rovable privacy鈥�.
Records of internet searches made on websites such as Google and AOL are hugely useful to software engineers trying to improve search technology. Such data also give social scientists a valuable window on our largely uninhibited digital search behaviours. The problem is that such information can easily identify individuals who have carried out the searches, breaching their privacy.
Until now, search engines鈥� attempts to anonymise this data have proved somewhat inept. In one case, before making the data public. But some queries proved so specific 鈥� such as people searching their own names and social security numbers 鈥� that reporters from The New York Times were able to use them to track down one individual.
Advertisement
鈥淧eople search for their own names and social security numbers, which renders them identifiable鈥�
Now a team of researchers at Microsoft鈥檚 search technology lab 鈥� Krishnaram Kenthapadi, Nina Mishra, Alex Ntoulas and Aleksandra Korolova 鈥� have developed a safer way to release search query logs. The idea is to publish data associated only with the most popular queries, so that rarely performed searches cannot be used to identify people with special, or even downright peculiar, interests.
Another way to 鈥渃rack鈥� anonymous data sets is to see how they compare to known data, a bit like identifying countries on an unlabelled map by comparing their shapes to a labelled map. To prevent this, the team inserts noise into the data by adding digits to the figures it contains. 鈥淎dding the noise gives provable privacy,鈥� says Korolova. And the amount of added noise defines the level of privacy that can be guaranteed. The work will be presented at the in Madrid, Spain, in April.
Is the resulting data set still useful to researchers, though? 鈥淚t depends what you use it for,鈥� says Korolova. The team has tested the technique by checking how search patterns reveal a ranking of people鈥檚 phobias. They found a similar set of rankings regardless of whether they used the privacy technique or not. 鈥淲e think we have a balance between a guarantee of privacy and the utility of the data sets,鈥� says Korolova.
Ian Brown, a researcher at the Oxford Internet Institute in the UK, is impressed: 鈥淭his shows privacy can be preserved in many types of systems without having to trade it against functionality.鈥�
