Contributions à la conception de systèmes de raisonnement à partir de cas. (Contributions to the design of case-based reasoning systems)

This document presents some researches centered on the issue of the design of a case-based reasoning (CBR) system, particularly of the steps of retrieval of a source case similar to the target problem and of adaptation of this case to solve this problem. First, an approach of retrieval based on the hierarchical organization of cases is described. It is composed of three procedures, the strong, fuzzy and smooth hierarchical classifications, and is correct in the sense that any retrieved case can be adapted to solve the target problem. This approach, in order to be applied, requires some adaptation knowledge. This kind of knowledge is studied under three viewpoints: general approaches of adaptation, adaptation knowledge acquisition and “adaptation catalogs” that are intended to be reusable from a CBR application to another. Finally, since a CBR system is a knowledge-based system, work on management of this kind of knowledge is described, in particular, for its representation. All this work comes for a big part from two projects: TICCO in the domain of synthesis planning in organic chemistry and KASIMIR in the domain of decision support in oncology.