Fish and Shrink. A Next Step Towards Efficient Case Retrieval in Large-Scale Case Bases

Keywords: Case-Based Reasoning, case retrieval, case representation This paper deals with the retrieval of useful cases in case-based reasoning. It focuses on the questions of what " useful " could mean and how the search for useful cases can be organized. We present the new search algorithm Fish and Shrink that is able to search quickly through the case base, even if the aspects that define usefulness are spontaneously combined at query time. We compare Fish and Shrink to other algorithms and show that most of them make an implicit closed world assumption. We finally refer to a realization of the presented idea in the context of the prototype of the FABEL-Project 1. The scenery is as follows. Previously collected cases are stored in a large scaled case base. An expert describes his problem and gives the aspects in which the requested case should be similar. The similarity measure thus given spontaneously shall now be used to explore the case base within a short time, shall present a required number of cases and make sure that none of the other cases is more similar. The question is now how to prepare the previously collected cases and how to define a retrieval algorithm which is able to deal with spontaneously user-defined similarity measures. 1 Motivation Suppose two cases in your domain have just been considered similar with respect to a certain aspect, and just a few minutes later they have to be regarded as to be considerably dissimilar because another aspect is now being regarded. If that cannot happen in your domain, then a lot of approaches making use of static similarities are available to organize your case base and it would be better to take one of them to realize your retrieval task. None of them will fit the requirements if similarity between two cases depends on the context in which retrieval is evoked. The reason is that once built up,