Exploiting similarity for supporting data analysis and problem solving

Abstract

Case-based reasoning relies on the hypothesis that “similar problems have similar solutions,” which seems to apply, in a certain sense, to a large range of applications. In order to be generally applicable and useful for problem solving, however, this hypothesis and the corresponding process of case-based inference have to be formalized adequately. This paper provides a formalization which makes the “similarity structure” of a system accessible for reasoning and problem solving. A corresponding (constraint-based) approach to case-based inference exploits this structure in a way which allows for deriving a similarity-based prediction of the solution to a target problem in form of a set of possible candidates (supplemented with a level of confidence.)