Journals Proceedings

Abstract

Textual case-based reasoning (TCBR) solves new problems by reusing previous similar problem-solving experiences documented as text. During reuse, TCBR identifies reusable textual constructs in the retrieved solution content and differentiates from the rest that need revision. However, reuse is heavily influenced by the quality of retrieval since TCBR attempts to adapt retrieved cases to solve a new problem. In scenarios where only the most similar case is adapted during reuse, such best match case might not necessarily be the easiest to adapt. We introduce a technique called Reuse Guided Retrieval to determine a specific similar case whose solution is best adaptable to solve a new query. A reuse metric is also proposed which encodes how easily reusable or adaptable the solution from a particular nearest neighbour is to a query. Experiments on two datasets from the domains of weather forecast revision and health & safety incident reporting indicate that our technique was more effective than a baseline retrieval which always chooses its best match using only the retrieval similarity between the cases and a query.