Strong, Fuzzy and Smooth Hierarchical Classification for Case-Based Problem Solving

This paper explains how case-based problem solving can have benefit from a hierarchical organisation of problems based on a generality relation. Three adaptation-guided retrieval processes are described. The strong classification in a problem hierarchy is a classical deductive process. It is based on the generality relation between problems which organises the hierarchy. The fuzzy classification is a fuzzification of the strong classification. It is based on a fuzzy generality relation between problems, which can be seen as a nonsymmetrical similarity measure. The smooth classification extends the fuzzy classification: it is also based on a similarity or dissimilarity measure but takes into account problem and solution adaptation knowledge. These processes have been successfully implemented in two case-based reasoning systems: RESYN/CBR in the domain of organic synthesis and KASIMIR/CBR in the domain of cancer treatment.