Extending the indexing vocabulary of case based reasoning with task specific features

Abstract

One of the central issues in case-based reasoning is the choice of an indexical vocabulary that allows for efficient retrieval of experiential knowledge from memory. The authors propose a new indexical vocabulary based on features of abstract problem types. The rationale for such a vocabulary was supported by empirical data gathered in the domain of alloantibody identification, and by the interpretation of this data in terms of the computational complexity of abductive reasoning. A case-based reasoning model of abduction is proposed that integrates both a domain and a problem type specific indexical vocabulary. The advantages of this model in terms of memory retrieval, case adaptation and problem solving are discussed.<>