- From: George Meditskos <gmeditsk@gmail.com>
- Date: Tue, 20 Apr 2010 19:01:08 +0300
- To: semantic-web@w3.org
- Message-ID: <g2pa838de391004200901pb21c5cb5zbda802a56a32ac78@mail.gmail.com>
The OWL2-RL semantics [1] can be implemented both with forward-chaining (e.g. RETE) and with backward-chaining rules. It depends on the type of reasoning someone wants. For example, in a forward-chaining implementation, there is a complete materialization of the semantics in the KB of the rule engine in advance, whereas in a backward-chaining implementation the reasoning is performed at runtime upon a request (query). A good example is the Jena API [2] that provides both a forward- and a backward-chaining implementation of the pd* entailments (it provides also a hybrid rule engine that combines forward and backward chaining rules). Another idea is to separate the reasoning on the ontology schema (TBox) from the reasoning on the instances (ABox). The former can be performed using a DL reasoner (e.g. Pellet [3]) and the latter with a rule engine (e.g. Jena). Such an implementation can take advantage of the TBox reasoning capabilities of DL reasoners, which are quite efficient in computing the terminological semantics, and the reasoning scalability that a rule engine can offer in large extensional knowledge bases. This is the idea of DLEJena [4][5] that uses Pellet for TBox reasoning and the Jena's forward-chaining rule engine in order to run a set of domain-dependent forward-chaining entailment rules that are generated at runtime, based on the Pellet's KB. Such a combination usually results in better reasoning performance than of implementing directly the semantics using only rules. Another interesting issue is whether an implementation invalidates the conformance requirements for OWL 2 RL or not (Theorem PR1 in [1]). In most practical cases, the classes of an ontology are not treated as instances at the same time and the reasoning is performed under the OWL 2 direct semantics. Such an implementation is not a conformant one, since OWL2 RL does not impose such a restriction, but it usually results in better reasoning performance, since some inferences are omitted. George [1] http://www.w3.org/TR/owl2-profiles/#Reasoning_in_OWL_2_RL_and_RDF_Graphs_using_Rules [2] http://jena.sourceforge.net/ [3] http://clarkparsia.com/pellet/ [4] http://lpis.csd.auth.gr/systems/DLEJena/ [5] http://dx.doi.org/10.1016/j.websem.2009.11.001
Received on Tuesday, 20 April 2010 18:08:29 UTC