Abstract

When reasoning with description logic (DL) knowledge bases (KBs) which contain a large number of axioms, performance is the key concern in real applications. To improve the performance, axiom absorption has been a central research issue in DL KBs. Well-known algorithms for axiom absorption, however, still heavily depend on the order and the format of the axioms occurring in the target KB. In addition, in many cases, there exist some restrictions in these algorithms which prevent axioms from being absorbed. Both the characteristics and the design of absorption algorithms for optimal reasoning are still open problems. In this thesis, we first seek to improve our theoretical understanding about the axiom absorption techniques including some related techniques such as simplification and normalization. Then we propose a criterion for the "best" absorption against experimental experience. Based on this criterion, we develop some new algorithms to absorb axioms in a KB to ameliorate the reasoning performance. The experimental tests we conducted are mostly based on synthetic benchmarks derived from common cases will occur in real KBs. The experimental evaluation demonstrates a significant runtime improvement.