Abstract: Traditional CBR system is planar architecture. With the system’s running and learning, case base will become larger and larger. When the numbers of cases surpass some boundary, there will be swarming problem. To settle this problem, this paper presents a CBR system based on quotient space model, which is hierarchical architecture. It combines active learning with lazy learning in system running phase. Experimental results show that based on this method, the system performance is greatly increased.

Key words: Quotient space theory, Information granularity, Hierarchy, Case-based reasoning

摘要： 传统的CBR系统采用平面结构，系统在运行过程中不断学习，范例库将变得越来越大，当范例数超过某一预设的上界时，就会出现“沼泽问题”。为了解决这个问题，该文提出了基于商空间模型的CBR系统，采用分层递阶的立体结构，在运行阶段将惰性学习算法与积极学习算法相结合。实验表明利用本方法构造的CBR系统实现E-mail分类预测时，系统的性能和有效性都得到了很大的提高。

关键词: 商空间理论, 信息粒度, 分层递阶结构, 范例推理