Formalizing CBR

ANALOG Project

Formalizing
Case-based
Reasoning

A Logical Approach to Case-Based Reasoning Using Fuzzy Similarity Relations

To be published in Information Sciences Journal

Available online

Enric Plaza F. Esteva, P. Garcia, L. Godo, R. López de Màntaras

Abstract

This article approaches the formalization of inference in Case-based Reasoning (CBR) systems. CBR systems infer solutions of new problems on the basis of a precedent case that is, to some extent, similar to the current problem. Using the logics developed for similarity-based inference we characterize CBR systems defining what we call the Precedent-based Plausible Reasoning PPR model. This model is based on the graded consequence relations named approximation entailment and proximity entailment. A modal interpretation is provided for the precedent-based inference where the plausibility is given by the graded possibility operator Diamond _alfa. The PPR model shows that both knowledge-intensive CBR systems and nearest neighbor algorithms share a common core formalism and that their difference is on whether or not (respectively) they use a general theory in addition to the precedent cases.

Bias in CBR Systems

In preparation

CONTACT Enric Plaza · enric@iiia.csic.es
· URL www.iiia.csic.es/Projects/analog/analog.html
· Phone +34 3 5809570 · Fax +34 3 5809661
· IIIA-CSIC, Campus UAB, 08193 Bellaterra, Catalonia, EU.