Ontology-Based Data Access Using Views

The OWL 2 QL profile, which is based on DL-LiteR, has been designed so that query answering is possible using relational database technology via query rewriting. Unfortunately, given a query Q posed in terms of an OWL 2 QL ontology O, the size of the rewritten query, Qo, which can be evaluated directly on the relational database, is worst case exponential w.r.t the size of Q and O [1]. This means that the computation and evaluation of Qo can be costly. Recent research focuses on creating rewriting algorithms that generatesQo with a smaller size [3]. In this paper, we propose a new approach to answering SPARQL queries on OWL 2 QL ontologies over existing relationally stored data. Our proposal is to replace answering queries via query rewriting with answering queries using views. Our position is that SQL infrastructure can be leveraged in order to support effective SPARQL query answering on OWL 2 QL ontologies over existing relationally stored data. We present preliminary results that support our position. Our position is inspired by our previous work on Ultrawrap [5], a system that can execute SPARQL queries at almost equivalent execution speed as its semantically equivalent SQL queries. Previous experimental studies demonstrated that existing approaches were several magnitudes slower. Our main insight was to represent the relational data as RDF triples using views. SPARQL queries are syntactically translated to SQL queries which operate on the views. We observed that two relational optimizations are needed in order for relational database to effectively execute SPARQL queries: detection of unsatisfiable conditions and self-join elimination. Given this past history, we ask ourselves if we can apply this same approach to answering SPARQL queries on OWL 2 QL ontologies. We first present the status quo with a running example adapted from [3]. We then present our proposed approach and show initial results that support our position. Finally, we present the open issues and our next steps.