Many companies’ privacy policies state they can only release customer data once personal identifiable information has been removed; however it has been shown by Narayanan and Shmatikov (2008) and reinforced in this paper that removal of personal identifiable information is not enough to anonymize datasets. Herein we describe a method for deanonymizing the Netflix Prize dataset users using publicly available Amazon review data [3], [4]. Based on the matching Amazon user profile, we can then discover more information about the supposedly anonymous Netflix user, including the user’s full name and shopping habits. Even when datasets are cleaned and perturbed to protect user privacy, because of the sheer quantity of information publicly available through the Internet, it is difficult for individuals or companies like Netflix to guarantee that the data they release will not violate the privacy and anonymity of their users.
[1]
Julian J. McAuley,et al.
Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering
,
2016,
WWW.
[2]
Vitaly Shmatikov,et al.
Robust De-anonymization of Large Sparse Datasets
,
2008,
2008 IEEE Symposium on Security and Privacy (sp 2008).
[3]
Anton van den Hengel,et al.
Image-Based Recommendations on Styles and Substitutes
,
2015,
SIGIR.
[4]
Daniel C. Barth-Jones,et al.
The 'Re-Identification' of Governor William Weld's Medical Information: A Critical Re-Examination of Health Data Identification Risks and Privacy Protections, Then and Now
,
2012
.