OBDA Using RL Reasoners and Repairing

ABSTRACT. In previous work it has been shown how a SROIQ ontology O can be `repaired' for an RL system ans---that is, how we can compute a set of axioms R that is independent from the data and such that ans that is generally incomplete for O becomes complete for all SPARQL queries when used with O and R. However, the initial implementation and experiments were very preliminary and hence it is currently unclear whether the approach can be applied to large and complex ontologies. Moreover, the approach could not support non-SPARQL CQs. In the current paper we thoroughly investigate repairing as an approach to scalable (and complete) ontology-based data access. First, we present several non-trivial optimisations to the first prototype. Second, we show how (arbitrary) conjunctive queries can be supported by integrating well-known query rewriting techniques with RL systems via repairing. Third, we perform an extensive experimental evaluation obtaining encouraging results. In more detail, our results show that we can compute repairs even for very large real-world ontologies in a reasonable amount of time, that the performance overhead introduced by repairing is negligible in small to medium sized ontologies and noticeable but manageable in large and complex one, and that our approach to handling non-SPARQL CQs can very efficiently compute the correct answers for real-world challenging TBoxes.