A Similarity Measure Based on Extension Distance and Its Application in CBR

Abstract

Case-based reasoning (CBR) is an effective method that integrates reasoning methodology and represents related knowledge in a domain. The success of a CBR system largely depends on case retrieval, and the similarity and determination of weight for each case features have a significant influence on the efficiency and accuracy of case retrieval. The aim of the research is to improve the efficiency and accuracy of case retrieval. Analyzing the deficiency of similarity measures based on the classical distance, different similarity measures are proposed for different kinds of attribute values based on the extension distance, especially the similarity model between numerical and set considered the customer’s preference. The standard deviation related with the similarity is introduced to distribute the dynamic attribute’s weights which also considered the customer’s interest, but not the traditional methods that the weight is a constant if determined. The presented methods will enable the system to retrieve the more similar case correctly so that reducing case adaptation. In this study, an electric drill is used as a case to verify the usefulness and effectiveness of the similarity measurements and weight assignments. It is demonstrated that this method is more beneficial to case retrieval compared with other methods.