Nearest-neighbor matching functions are commonly use din case-based reasoning systems to assess the overall similarity betweena an old case and the tagert problem by a weighted sum of individual similarity measures along case attribute. It is discussed in this paper that the linear form of weighted sum in the nearest-neighbor matching function may not well address the nonlinear effects of attributes toward the overall similarity, and the imprecise nature of weights associated with attributes may fail to produce robust results in care retrieval. To remedy these two deficiencies, a generalized nearest-neighbor matching function based on a fuzzy inference system is proposed. This approach also has the advantage of incorporating the user's knowledge in similarity assessment. An application to the case-based retrieval for new product design illustrates the proposed approach. Comparisons between the proposed approach and the traditional nearest-neighbor matching functon are carried out to show the robustness of the proposed approach.
關聯:
Mathematical and Computer Modeling 38(3-4), pp.385-394
