Abstract: The evaluation index of conventional case base maintenance method was easily interfered by subjective experiences, and could not reflect the problem solving ability of case base. Aiming at this problem, a new maintenance method based on comprehensive correlation was proposed. By analyzing coverage and reachable measure, comprehensive correlation evaluation model was built. The outlier data of case base were analyzed by using Clustering LARge Application(CLARA). Footprints were discovered in the correlated set by means of measurement computation, and a subset was found with the same competence as the primitive case base. Two groups of experiments were used to show the feasibilty of method.

Key words: case-based reasoning, case base maintenance, evaluation index, clustering, outlier data

摘要： 以往案例库维护方法定义的能力评价指标受主观因素影响，且不能从整体上反映案例库解决问题的能力，为此提出一种基于综合相关性的维护方法。以覆盖性和可达性为基础，建立综合相关性评价模型；利用CLARA进行案例库的离群数据分析，依据综合相关性度量发现相关子集的脚标，得到原案例库的能力等价集。通过两组实验说明了方法的可行性。

关键词: 基于案例推理, 案例库维护, 评价指标, 聚类, 离群数据