A solution space approach to handling constraints in case-based reasoning: the unified problem

This paper aims to answer the following research question: “Can the standard CBR models be generalised to a unified (problem : solution) space to allow flexible query modes?“ In the standard Case-Based Reasoning (CBR) model, a case is represented as a <problem, solution> pair. The problem space and solution space are treated as separate, and nearest neighbours are retrieved using a metric defined on the problem space. The standard method applies to domains where the similarity assumption is valid: that cases which are near in the problem space are also near in the solution space. This presumes that a metric is also defined in the solution space. In this paper a generalisation of the standard CBR retrieval method, which integrates solution space and retrieval space into a single query space is proposed. The retrieval method is proposed by means of the concept of nearest neighbours to a constraint region. It is shown that the standard CBR retrieval is a special case of this more general model. The advantages of the general model are explained in connection with its more general query modes. Whereas the standard model is only queryable by specification of problem “inputs”, the more general model is capable of retrieving general queries on the unified problem-solution space. The advantages of this flexible query form are explained in the paper, by means of a variety of illustrative examples.