A method of similarity metrics for structured representations

摘要

Case-based reasoning (CBR) is a topic that becomes more and more important and recently has been used for planning, design, etc., of applications in industry and business domaines. In case-based reasoning, a problem is solved by recognizing its similarity to a known problem (i.e. a case) and then adapting the corresponding solution to solve the new problem. Consequently, similarity metrics play a central role in CBR. This paper proposes a method of similarity metrics based on cases being represented by structured representations. First, we define the degree of similarity between two cases, which is used as the retrieving criterion, and the similar correspondence which presents the information of case adaptation. Then we derive some theorems for their computation. Because this computation problem can be translated to a combinational optimization problem, genetic algorithms, which can find near optimal solutions quickly, can be applied. We develop a GA for our problem. Finally, we show the results of our simulation and compare it with other methods.