Paper: cs.CR/0610105
Date: Wed, 18 Oct 2006 06:03:41 GMT (128kb)
Title: How To Break Anonymity of the Netflix Prize Dataset
Authors: Arvind Narayanan and Vitaly Shmatikov
Subj-class: Cryptography and Security; Databases
\\
As part of the Netflix Prize contest, Netflix recently released a dataset
containing movie ratings of a significant fraction of their subscribers. The
dataset is intended to be anonymous, and all customer identifying information
has been removed. We demonstrate that an attacker who knows only a little bit
about an individual subscriber can easily identify this subscriber's record if
it is present in the dataset, or, at the very least, identify a small set of
records which include the subscriber's record.
A successful deanonymization of the Netflix dataset has possible implications
for the Netflix Prize, which promises $1 million for a 10% improvement in the
quality of Netflix movie recommendations. Given movie ratings of Netflix users
who have made their ratings public, or perhaps obtained from public sources
such as the Internet Movie Database (IMDb), a contestant may be able to
identify their records in the Netflix dataset (if they are among the Netflix
subscribers whose anonymized records have been released). With the complete
knowledge of a subscriber's IMDb ratings, it becomes much easier to "predict"
how he or she rated any given movie on Netflix.
\\ ( http://arXiv.org/abs/cs/0610105 , 128kb)
--
__Pascal Bourguignon__ http://www.informatimago.com/
"Debugging? Klingons do not debug! Our software does not coddle the
weak."