undefined | Better HN

0 pointssitkack10y ago0 comments

The data cannot for all intents and purposes be anonymized and still be useful. I have seen airsage data, they give the user a new unique identifier but it is consistent over a 30 day window. Very easy to deanonymize, the unique identifiers are only opaque to people who can't do math.

0 comments

3 comments · 1 top-level

e12e10y ago· 2 in thread

Yeah, data anonymization doesn't really work, except perhaps for highly aggregated data (eg: percentage of people out of work in a nation). It does provide a small barrier in the form that "if you don't look, you won't see" -- so it might prevent some inadvertent leaks of private information.

But it should rarely be considered a "security boundary" so to speak.

aianus10y ago

Why not? It's not hard to set up streaming filters that count how many people visit a certain mall in realtime without ever persisting the raw location stream.

e12e10y ago

Right. But when the data you have is detailed logs of individual movements, second by second, meter by meter - stepping down to "number of unique IMEIs in a 100 square km area" is moving to "highly aggregated data". If that is what is seen as anonymization, the yeah. But then you've pretty much reduced a dataset to information that answers a specific question. Typically anonymization is (AFIK/IMNHO) taken to mean taka a dataset that can answer some questions, and transform it to a dataset that can still answer some questions we haven't come up with, but avoids answering questions that lead to identifying individuals. Eg: the problem with the NYC taxi driver data:

http://arstechnica.com/tech-policy/2014/06/poorly-anonymized...

j / k navigate · click thread line to collapse