High Performance Reasoning with Very Large Knowledge Bases: A Practical Case Study

Abstract

We present an empirical analysis of optimization techniques devised to speed up the so-called TBox classification supported by description logic systems which have to deal with very large knowledge bases (e.g. containing more than 100,000 concept introduction axioms). These techniques are integrated into the RACE architecture which implements a TBox and ABox reasoner for the description logic ALCN HR+ . The described techniques consist of adaptions of previously known as well as new optimization techniques for efficiently coping with these kinds of very large knowledge bases. The empirical results presented in this paper are based on experiences with an ontology for the Unified Medical Language System and demonstrate a considerable runtime improvement. They also indicate that appropriate description logic systems based on sound and complete algorithms can be particularly useful for very large knowledge bases.