Graph Automorphism-Based, Semantics-Preserving Security for the Resource Description Framework (RDF).

We address security in the context of the Resource Description Framework (RDF), a graph-like data model for the web. One of RDF's compelling features is a precise, model-theoretic semantics. We first propose a threat model, and under it, observe that the technical challenge is really in hiding information that may be revealed by the structure of an RDF graph. We choose two quantitative, unconditional notions for securing graph-structure from the literature that address the threat model, and adapt them for RDF. We then consider the problem of devising algorithms for achieving a certain level of security while preserving the semantics of the input RDF graph. We observe that there are operations we can perform on an RDF graph that both provide such security and preserve semantics. We observe, further, that there is a natural way to quantify information-loss under these operations, and that there appears to be a natural trade-off between security and information-quality. We study this trade-off and establish fundamental results. We show that the RDF graphs that result from applying the operations induce a lattice that leads to a natural quantification of information-quality. We show also that achieving a certain level of security while retaining a certain level of information-quality is NP-complete under polynomial-time Turing reductions. Finally, towards an empirical assessment, we discuss our design and implementation of a reduction to CNF-SAT, and empirical results for two classes of RDF graphs. In summary, our work makes fundamental and practical contributions to semantics-preserving security for RDF.