Automated Construction of Knowledge-Bases from Examples

The process of knowledge acquisition has long been regarded as the bottleneck in expert systems development. In this paper, a concept induction methodology for automated construction of knowledge-bases is presented and its use for knowledge acquisition is discussed. The applicability of such a tool to build expert systems is demonstrated using CONIS in some selected experiments. CONIS is a concept induction system that infers concept description from sample instances of the concept. We have also compared CONIS with conventional statistical techniques in solving classification problems. The results suggest that concept induction could become be a viable tool to automate the process of knowledge acquisition. By shortening the development cycle, domains that were once too volatile for expert systems application would become feasible using such an automated aid.