DOI:
Abstract:
It is essential for case-based reasoning (CBR) systems to access trnly relevant cases efficiently. Similarity assessment adopted by many CBR systems needs performance improvement, especially if the case library of a CBR system consists of a database in the target domain which is not equipped with suitable indexing for the CBR system. In consideration of this problem, we are exploring use of strict matching, a conventional information retrieval technology, to screen out irrelevant cases effectively and efficiently before similarity assessment. In this paper, we first clarify the need for combining strict matching and similarity assessment by analyzing two CBR systems we developed: Sales Estimation System and Motor Design Support System. Then, we propose a method for combining the two retrieval techniques, Similarity Assessment with Strict Screening, which first performs strict matching to reduce the size of a set of candidate cases subject to similarit~t assessment. Central to the method is an algorithm which is complete in that strict matching does not screen out any relevant case which similarity assessment would select.