Experimental study of an evaluation function for cases imperfectly explained

Abstract

This paper describes an evaluation function which deals with similarities and dissimilarities in cases imperfectly explained. Our explanation- based evaluation function represents an alternative to other approaches that use case explanations for selection and retrieval of past cases. This function is used by a diagnosis system called RECIDEclinic. We present the experimental results obtained in the domain of neurologic diseases. These results illustrate the role of similarity and dissimilarity terms as they are defined within our framework.