Provable De-anonymization of Large Datasets with Sparse Dimensions

Abstract

There is a significant body of empirical work on statistical de-anonymization attacks against databases containing micro-data about individuals, e.g., their preferences, movie ratings, or transaction data. Our goal is to analytically explain why such attacks work. Specifically, we analyze a variant of the Narayanan-Shmatikov algorithm that was used to effectively de-anonymize the Netflix database of movie ratings. We prove theorems characterizing mathematical properties of the database and the auxiliary information available to the adversary that enable two classes of privacy attacks. In the first attack, the adversary successfully identifies the individual about whom she possesses auxiliary information (an isolation attack). In the second attack, the adversary learns additional information about the individual, although she may not be able to uniquely identify him (an information amplification attack). We demonstrate the applicability of the analytical results by empirically verifying that the mathematical properties assumed of the database are actually true for a significant fraction of the records in the Netflix movie ratings database, which contains ratings from about 500,000 users.

On the feasibility of user de-anonymization from shared mobile sensor data

“…Our results build on existing frameworks that exploit data sparsity [6,12,8,19]. However, we examine a new type of data and de-anonymization scenarios previously not studied.…”

“…Researchers have discovered that maintaining user anonymity within sparse datasets is surprisingly difficult [6,12,8,19]. Within this research community, the term sparsity refers to datasets in which an individual user or identity can be distinguished from others in a dataset by only a few select rarely occurring user attributes.…”

A Literature Survey and Classifications on Data Deanonymisation

“…There are also more studies in the literature that researched deanonymisation [21][22][23][24][25][26][27] but these are not included in our survey as they are for analysis and evaluation purposes.…”

On the use of decentralization to enable privacy in web-scale recommendation services

“…Second, we observe in this work that the primary reason for privacy leakage in auxiliary information attacks [11,26] is a centralized untrusted entity gathering detailed consumption about a pseudonymous end-user, even if the user's consumption is hidden within a group. A single untrusted aggregator possessing the scrambled consumption of all groups can mount attacks by first identifying the group that a user belongs to by exploiting auxiliary information, and can then infer sensitive attributes of the user if the group happens to predominantly revolve around sensitive topics (l-diversity [22] attacks).…”

“…This is because revealing multiple interests of a pseudonymous user is open to attacks based on auxiliary information, wherein an individual can be uniquely identified using a combination of few consumed items or interests (see Netflix attack [11,26]). …”