Cbr Textuality

Current textual CBR research focuses on generating knowledge-rich representations for working with document-cases. However, there are many \weakly-textual" contexts in which textual information plays an important , but ancillary role in case composition and reasoning. For retrieving weakly-textual \semi-structured" cases that contain one or more textual features, it is desirable to measure the similarity of the textual components in a simple manner that is easily integrated with standard, non-textual similarity metrics, such as nearest-neighbor methods. This paper describes an investigation of the issues involved in weakly-textual CBR, which has been motivated by work on the DRAMA system for aerospace design support. It describes a characterization of textual CBR along a continuum of textual importance in case composition and reasoning, that places weakly-textual CBR in relation to current practice and motivates a focus on weakly-textual methods. The paper goes on to discuss methods, which are currently being incorporated into DRAMA, that integrate simple information retrieval techniques into standard CBR similarity metrics for weakly-textual contexts, and it concludes with a discussion of related work and future directions.