Mining Adaptation Rules from Cases in CBR Systems

Abstract

Case based reasoning (CBR) is a well known framework to cope with ill-structured tasks, where no enough domain knowledge is available [2]. The main objective of CBR is to address the knowledge acquisition bottleneck. Namely, in CBR the reasoner does not make effort to build an abstract model for domain knowledge to solve the problem, instead, during the problem solving, it relies on the past similar cases, and attempts to find the appropriate solution for the problem at hand, by modifying the past similar solutions. However, CBR systems also require substantial knowledge acquisition effort (e.g. acquiring cases, case vocabulary, retrieval knowledge, adaptation knowledge [5]). This knowledge traditionally is derived from a domain expert. Accordingly, although the expert can not propose an abstract model to support the domain, s/he attempts to define some regularities in the domain, that makes it possible to reason with the cases. In fact, the knowledge necessary to solve the problem has been depicted in the cases. The expert defines the skill of using the right case in the right place (similarity), and also defines a formalism to modify the old cases such that address the new problem (adaptation). Defining this knowledge, if precision matters needs an accurate model of domain, proposed by expert (which is normally not available in CBR domains), and if precision is not assumed as a requirement, may fail to demonstrate acceptable reasoning skill. However, since in CBR there is no well-defined background knowledge, defining this knowledge by expert, constrains the system with a static incomplete reasoning skill, that does not evolve with new experiences, even though this skill can be acquired through the new cases. Furthermore, adaptation is a compensatory part of CBR [1]. Adaptation process can compensate the shortcomings in other steps of reasoning process, by appropriately modifying a weak match to obtain a reasonable solution. Accordingly automating the process of acquiring adaptation knowledge, in this work, has been proposed in the means of increasing the reasoning skill.