Faster OWL Using Split Programs

Knowledge representation and reasoning on the Semantic Web is done by means of ontologies. While the quest for suitable ontology languages is still ongoing, OWL [5] has been established as a core standard. It comes in three flavours, as OWL Full, OWL DL and OWL Lite, where OWL Full contains OWL DL, which in turn contains OWL Lite. The latter two coincide semantically with certain description logics and can thus be considered fragments of first-order predicate logic. OWL ontologies can be understood to consist of two parts, one intensional, the other extensional. In description logics terminology, the intensional part consists of a TBox and an RBox, and contains knowledge about concepts (called classes) and complex relations between them (called roles). The extensional part consists of an ABox, and contains knowledge about entities and how they relate to the classes and roles from the intensional part. The Semantic Web envisions a distributed knowledge source, built from OWL ontologies and intertwining the knowledge like the Web interconnects websites today. With an estimated 25million active websites today and correspondingly more webpages, it is apparent that reasoning on the Semantic Web will have to deal with very large ABoxes. Complexity of ABox reasoning — also called data complexity—measures complexity in terms of ABox size only, while considering the intensional part of the ontology to be of constant size. For the different OWL variants, data complexity is at least NP-hard, which indicates that it will not scale well in general. Therefore, methods are sought to cope with large ABoxes in an approximate manner. The idea is to use quick heuristic reasoning when time constraints are more important than the correctness of the answers. A typical use case is online question answering, where it is more important to give rough answers quickly than to have precise responses at the cost of long delays.