一种基于实例推理的概念学习方法

A Concept Learning Method in Case-based Reasoning

特征结构是一种新的实例表示方法,结合本体提供的术语和类别层次关系特别易于表达复杂的关系结构型实例,首先讨论了一种基于特征结构实例推理的概念学习方法:C-LID算法,在提出基于特征结构实例推理解决概念学习问题的相似原则和语义包含原则基础上分析了C-LID算法的缺点,进一步提出了基于豪斯多夫距离和K-S相似度的消极概念学习方法R-LID,一方面R-LID是C-LID算法在相似原则下的扩展,另一方面R-LID避免了最多最优偏置下近邻选取不当造成的误差。将R-LID用于化合物致癌等级划分的开放问题上,结果表明R-LID算法具有更好的性能。

Feature term is a new case representation method,combined with the partial orders in taxonomic ontology, which is able to encode complicated relational cases.Firstly,a concept learning method in feature term Case-based Reasoning C-LID is introduced and two basic principles are proposed.Secondly,a lazy learning algorithm R-LID based on Hausdorff distance and K-S similarity is put forward to ameliorate C-LID in two aspects： （1）to define a extended approach to C-LID; （2）to avoid the negative effect under majority rule.Experiments on the Toxicology dataset show that R-LID can do better than H-S,LID and C-LID.