Improving a distributed case-based reasoning system using introspective learning

This paper describes the improvements to a fielded case-based reasoning (CBR) system. The system, called "Really Cool Air" is a distributed application that supports engineering sales staff. The application operates on the world wide web and uses the XML standard as a communications protocol between client and server side Java applets. The paper briefly describes the distributed architecture of the application, the two case retrieval techniques used, and improvements made in three areas: software infrastructure, retrieval algorithm and most crucially the automatic adjustments of query relaxation parameters by an incremental learning algorithm. The results of the incremental learning algorithm are quantified.