Eric K. Jones and Aaron Roydhouse
We argue that the process of designing effective systems for case retrieval should not be divorced from the process of evaluation. Developing a suitable representation vocabulary for indices, and suitable mechanisms for case retrieval should proceed in an iterative fashion, guided by the joint application of sound AI principles, domain expertise, and empirical evaluation. We lay out the key components of a case retrieval system and suggest procedures for evaluating each component. We illustrate our claims using the Metvuw Workbench, a system for intelligent retrieval and display of historical meteorological data.