Reasoning with Cases Imperfectly Described and Explained

This paper introduces RECIDE, an implementation of our approach to case-based reasoning. A qualitative and a quantitative metric are used for case retrieval. RECIDE has a library of successful and failure cases. Generation of new solutions is driven by splitting and merging operations on successful cases. Failure cases are in the form of indivisible and incompatible cases and represent constraints on the application of splitting and merging operators. Both types of cases are acquired interactively during problem-solving. We present the algorithms for generation of new cases with a solution that potentially applies to the new problem. RECIDEpsy an application of RECIDE in the domain of psychology, is introduced in this paper.