Optimizing Retrieval in CBR by Introducing Solution Similarity

In Case-Based Reasoning it is of vital importance that one is able to define similarity measures that approximate sufficiently the utility of existing case knowledge with respect to current problem situations. Although in many CBR applications pure geometric distance measures lead to reasonable results, the integration of additional domain knowledge into the similarity measure usually improves the problem solving capabilities of CBR significantly. The aim of this paper is to present an approach that facilitates the definition of such knowledge intensive similarity measures. This approach bases on a machine learning strategy that utilizes known relations between the solution parts