Abstraction Refinement for Ontology Materialization Technical Report

ion Refinement for Ontology Materialization Technical Report Birte Glimm, Yevgeny Kazakov, Thorsten Liebig, Trung-Kien Tran, and Vincent Vialard 1 University of Ulm, Ulm, Germany, [email protected] 2 derivo GmbH, Ulm, Germany, @derivo.de Abstract. We present a new procedure for ontology materialization (computing all entailed instances of every atomic concept) in which reasoning over a large ABox is reduced to reasoning over a smaller “abstract” ABox. The abstract ABox is obtained as the result of a fixed-point computation involving two stages: 1) abstraction: partition the individuals into equivalence classes based on told information and use one representative individual per equivalence class, and 2) refinement: iteratively split (refine) the equivalence classes, when new assertions are derived that distinguish individuals within the same class. We prove that the method is complete for HornALCHOI ontologies, that is, all entailed instances will be derived once the fixed-point is reached. We implement the procedure in a new database-backed reasoning system and evaluate it empirically on existing ontologies with large ABoxes. We demonstrate that the obtained abstract ABoxes are significantly smaller than the original ones and can be computed with few refinement steps. We present a new procedure for ontology materialization (computing all entailed instances of every atomic concept) in which reasoning over a large ABox is reduced to reasoning over a smaller “abstract” ABox. The abstract ABox is obtained as the result of a fixed-point computation involving two stages: 1) abstraction: partition the individuals into equivalence classes based on told information and use one representative individual per equivalence class, and 2) refinement: iteratively split (refine) the equivalence classes, when new assertions are derived that distinguish individuals within the same class. We prove that the method is complete for HornALCHOI ontologies, that is, all entailed instances will be derived once the fixed-point is reached. We implement the procedure in a new database-backed reasoning system and evaluate it empirically on existing ontologies with large ABoxes. We demonstrate that the obtained abstract ABoxes are significantly smaller than the original ones and can be computed with few refinement steps.