Abstract

Description logic knowledge bases traditionally contain a set of axioms (Tbox) describing background knowledge. DL reasoners generally handle axioms by using the lazy unfolding technique, which reduces the nondeterminism introduced by axioms. Axiom absorption is an optimization technique that rewrites axioms into the unfoldable part of the Tbox suitable for lazy unfolding. Absorptions are generally employed in DL reasoners in a mostly uniform way regardless of the characteristics of an input knowledge base. Though there exist a number of absorptions, their overall effectiveness remains to be improved, especially when a large quantity of complex axioms are present in the knowledge bases, which is well beyond the capability of any single absorption technique. To ameliorate absorption techniques, this thesis presents a framework applying AI planning to axiom absorption. In this framework, a state space planner is used to encode state-of-the-art absorption techniques. Some designed heuristics concerning the characteristics of an input KB are utilized for the cost estimation during planning. The planner first applies appropriate absorptions to axioms, then it produces a solution with a minimized cost. Such a solution automatically organizes absorptions in a certain sequence to maximize the number of axioms for absorptions. Compared to a predetermined or fixed order of applying absorption techniques, the proposed framework benefits from the advantages to consider more absorption alternatives, which tends to be more flexible and effective.