ABSTRACT
Case-based reasoning (CBR) operates by remembering, adapting, executing, and storing episodes of problem-solving in a case library. Two requirements for CBR systems are (i) a set of retrieval rules and (ii) a set of adaptation functions. Current practice is to handcraft index rules and adaptation functions manually. This approach is time-consuming and lacks ability to grow beyond its initial implementation. In the paper we describe preliminary work on the development of ways to learn the rules and functions dynamically by “watching” a case library. The application domain is a communications network fault resolution system.
