Privacy-Preserving Ontology Publishing for $$\mathcal {EL} $$ Instance Stores

Computing Compliant Anonymisations of Quantified ABoxes w.r.t. $$\mathcal {EL} $$ Policies

“…These results improve on the ones in [3], where a severely restricted form of ABoxes, called instance stores, was investigated. The ABox in our example is not an instance store, due to the role assertion between the individuals d and g. Note that, even in this restricted case, the set of optimal compliant anonymisations may be exponentially large, which demonstrates that the exponential complexity of our algorithms cannot be avoided.…”

“…In general, this set contains exponentially many anonymisations that may be of exponential size. As already shown in [3] for the restricted case of an EL instance store, this exponential blowup cannot be avoided in the worst case, both regarding the number and the size of the anonymisations. These exponential lower bounds hold both for the case of classical entailment and of IQ-entailment (since for instance stores this does not make a difference).…”

“…Entailment of pp formulas corresponds to CQ containment, which is a well-known NP-complete problem [10]. 3 The relational datasets with labelled null values (which generalize RDF graphs) considered in [12,13] can also be viewed as pp formulas, where the quantified variables are the labelled null values.…”

“…In the controlled query evaluation framework, funded by the Deutsche Forschungsgemeinschaft (DFG) -430150274 the source data are left unchanged, but an additional layer, called censor, is introduced, which decides whether and how queries are answered [9,11,16]. In contrast, anonymisation approaches modify the source data in a minimal way such that secrets that should be preserved can no longer be derived [3,[12][13][14]. We use the approach for privacy-preserving publishing of linked data introduced in [12,13] as a starting point, where the information to be published is a relational dataset, possibly with (labelled) null values, and the privacy constraints (called policy) are formulated as conjunctive queries.…”

“…In [12,13] and [3], safety is introduced as a strengthening of compliance. Basically, safety means that the hidden facts should not be derivable even if additional compliant information is added.…”