N. Xiong and P. Funk (Sweden)
Similarity, utility, case based reasoning, classification.
Defining proper similarity functions is a key issue in the development of case-based reasoning systems. The success of problem solving by a system relies on its ability to retrieve the most usable or adaptable cases. A framework for utility oriented similarity modeling is proposed in this paper. The goal is to promote more accurate similarity assessments in accordance with utility. This goal can be achieved by adapting an underlying similarity model to mimic utility samples derived from the case library. A new structure of similarity is introduced which appears more suitable for utility approximation and also enables the encoding of single features impacts. The proposed approach has been applied to classification of real datasets and the obtained results from experiments seem promising.
Important Links:
Go Back
