Mark Winter and D. Sleeman
Abstract
Case-Based Reasoning (CBR) is a paradigm which makes use of similarity
to reuse previous problem solving incidents when solving a new
problem. These stored problem solving incidents are known as cases and
are stored in a case-base which acts as the main source of knowledge
in a CBR system. The construction of a knowledge base typically
involves three stages: eliciting the knowledge somehow, transforming
this into the appropriate representation and refining (correcting) any
errors in the resulting knowledge-base. Knowledge-base refinement
systems (partially) automate this final stage, but the majority have
dealt with rule-based representations. It is important for the
successful use of CBR systems that their case-bases are consistent. To
this end, this paper describes two knowledge refinement systems which
refine case knowledge. REFINER deals with cases which have been
classified by a domain expert and forms generalised descriptions of
each of the categories which are used to detect and help remove
inconsistent cases. The REFINER+ system addresses several shortcomings
of REFINER but shares the same overall aims.