Analysis of Topological Privacy Notions

The current state-of-the-art notions in statistical disclosure control face some problems in
real-world data privacy scenarios, by either making best-case assumptions or potentially
degrading the utility of the data. However, in 2017 Erdmann introduces two new syntactic
privacy notions for Relations: Attribute Privacy and Association Privacy. These notions
are defined over the topology of the released data. Notably, they do not include best-
case assumptions, and typically syntactic notions provide good data utility. It is thus
worthwhile to analyze their benefits and drawbacks in the context of existing notions.
Nevertheless, up to this point, no formal analysis has been performed, and no mechanism
has been given for these notions.
Therefore, in this work, we formally analyze the topological notions, providing theoretical
proofs about their protections, and compare them to the current state-of-the-art privacy
notions in statistical disclosure control. Further, we provide a mechanism to achieve
Attribute Privacy, whose utility we empirically analyze. Overall, we reveal multiple flaws
inherent to the topological notions, and come to the conclusion that they are not suited
... mehr
as a generalized replacement for the existing relational data state-of-the-art. Additionally,
as Association Privacy has the protection goal of hiding associations between people,
we provide an analysis of Association Privacy in the context of graph data privacy. We
find that the existing syntactic graph privacy notion of k-security provides stronger
protections.