Engineering optimisations in query rewriting for OBDA

Abstract

Ontology-based data access (OBDA) systems use ontologies to provide views over relational databases. Most of these systems work with ontologies implemented in description logic families of reduced expressiveness, what allows applying efficient query rewriting techniques for query answering. In this paper we describe a set of optimisations that are applicable with one of the most expressive families used in this context (tLrU-O ). Our resulting system exhibits a behaviour that is comparable to the one shown by systems that handle less expressive logics. INTRODUCTIONOntology Based Data Access (OBDA) consists on superimposing a conceptual layer as a view to an underlying information system, which abstracts away from how that information is maintained in the data layer and provides inference capabilities [6]. Expressiveness comes at a cost in terms of computational complexity, and thus OBDA has traditionally been an expensive feature only interesting when the challenges to tackle were on par with the cost of the solution. Several languages have been proposed as interesting compromise solutions between expressiveness and tractability for OBDA: the DL-Lite family [5], which includes DL-Lite core , DL-Lite^ and DL-Lite-ji; the QL profile of OWL2 [8]; some families in Datalog± [4]; and the tLrU-O family [14].In this paper we focus on ECHTO¬ and propose a series of optimisations that can be carried out during the rewriting process, which we have implemented in our system kyrie. This paper is structured as follows: In section 2 we provide some background on DL families and approaches for OBDA query rewriting. In section 3 we introduce the algorithms involved in our approach. In section 4 we describe our optimisations for the query rewriting process for tLtU-U . In section 5 we evaluate the results obtained. Finally, in section 6 we draw some conclusions from the work performed and outline some foreseen further improvements in the rewriting process. BACKGROUNDNow we briefly describe the most relevant logics used for OBDA, and review implementations that use these logics. They are presented here in a chronological order. OBDA-related logicsIn this section we cite logics with special relevance in OBDA.In the examples a refers to a constant (individual), Bi refer to basic concepts (classes) and R refers to roles (properties).• The DL-Lite family diverged into logics DL-Lite-ji and DL-Lite^, both extending DL-Lite core , where concept inclusions are restricted to B\ C B2 and B\ C ¬B2. DL-Lite-ji includes subsumption (ISA) and disjointness assertions between roles and DL-Lite^ includes functionality restrictions on roles. These logics are first-order reducible with a tractable complexity [5].• The OWL2 QL profile was inspired by the DL-Lite family and designed to keep the complexity of rewriting low, considering first-order rewritability. As a summary of a more extensive comparison [1], the main differences with DL-Lite are the lack of number restrictions, functionality constraints and keys. Among the constructs in OWL ...

Query Rewriting in RDF Stream Processing

“…Query Rewriting in ELHIO Query rewriting in our approach uses a nonrecursive ELHIO ontology to rewrite the conjunctive query q into a union of conjunctive queries (UCQ) q such that cert(q, O) = {α | q ∪ A |= q h (α)}. The details of the inference steps for the rewriting can be found in [18]. The transformation of the ontology to first order logic is performed according to the rules described in [21].…”

“…This prototype is available as open-source code 2 , and is based on two main components: (i) the query rewriter kyrie [18], which rewrites queries using ELHIO ontologies, and (ii) the RSP query engine CQELS [16]. The whole process is divided in a series of steps that are graphically summarized in Figure 1.…”

“…For instance, temperature observations are observed by a temperature sensor, a thermistor is a type of temperature sensor, etc 3 : Pre-processing. When this query is registered, the rewriting process is launched using the kyrie [18] module. kyrie uses ELHIO as the language for the ontologies, as it is one of the most expressive DLs that can be currently handled in query rewriting (see Section 6 for details).…”

“…For example, the following conjunctive query clauses can be obtained from the query in Listing 2: Saturation and expansion. These clauses with the conjunctive query form a logic program that is saturated by using resolution with free selection (RFS) and including a series of optimizations, as described in [18]. This results in a number of clauses that is usually smaller than those produced by other similar techniques (e.g.…”

“…Our approach demonstrates that these engines can be coupled with a query rewriter, and still be efficient for a large range of scenarios. Furthermore we implemented our solution, called StreamQR, extending the CQELS query processor with the kyrie [18] rewriting engine.…”