CBR applications running in real domains can easily reach thousands of cases, which are stored in the case library. Retrieval times can increase greatly if the retrieval algorithm can not cope with such an amount of cases. Redundancy can also be a problem, focusing retrieval alternatives in a very restricted search space. Basically, the system’s performance starts to degrade with the increase of the case-base size. Casebase maintenance allows CBR systems to deal with this problem, mainly through the use of case selection criteria. In this paper we present an experimental study about several case-base maintenance policies developed till now. We adapted and implemented these policies to a CBR system for software reuse and design, testing the applicability of these policies to cases with a complex representation (combination of tree and graph representations).
Paulo Gomes, Francisco C. Pereira, Paulo Paiva, Nu