Case-based reasoning systems routinely record the results of prior problem-solving, but not the provenance of new cases: the way in which the new cases were derived. This paper proposes the value of tracking provenance information in CBR, especially when timely feedback may not be available. It illustrates the use of provenance information with studies of the application of provenance information to guide case-base maintenance. Experiments with two data sets illustrate the benefit of using provenance to propagate maintenance and to target maintenance effort.
David B. Leake, Matthew Whitehead