Publications

Formal methods are a usual means to avoid errors or bugs in the development, adjustment and maintenance of both software and knowledge bases. This chapter provides a formal method to refine a knowledge base based on insides about its correctness derived from its use in practice. The objective of this refinement technique is to overcome particular invalidities revealed by the application of a case-oriented validation technology, i.e. it is some kind of "learning by examples". Approaches from AI or Data Mining to solve such problems are often not useful for a system refinement that aims at is an appropriate modeling of the domain knowledge in way humans would express that, too. Moreover, they often lead to a knowledge base which is difficult to to interpret, because it is too far from a natural way to express domain knowledge. The refinement process presented here is characterized by (1) using human expertise that also is a product of the validation technique and (2) keeping as much as possible of the original humanmade knowledge base. At least the second principle is pretty much adopted from Software Engineering. This chapter provides a brief introduction to AI rule base refinement approaches so far as well as an introduction to a validation and refinement framework for rulebased systems. It also states some basic principles for system refinement, which are adopted from Software Engineering. The next section introduces a refinement approach based on these principles. Moreover, it considers this approach from the perspective of the principles. Finally, some more general conclusions for the development, employment, and refinement of complex systems are drawn. The developed technology covers five steps : (1) test case generation, (2) test case experimentation, (3) evaluation, (4) validity assessment, and (5) system refinement. These steps can be performed iteratively, where the process can be conducted again after the improvements have been made.