基于等级相关的非分类案例检索

Non-Classification Case Retrieval Based on Rank Correlation

摘要: 现有的非分类CBR（case-based reasoning）系统的对齐度量需要设定阈值，为了克服此局限，提出使用等级相关来判断一个案例是否符合CBR假设，据此给出了非分类CBR系统的评价指标，对加拿大交通安全局的55个航空事故调查报告进行实验，结果表明，使用等级相关对齐进行5-NN案例检索比使用案例对齐（case alignment）和加权相关（weighted correlation）这两个对齐度量，正确率分别提高了10.91%和16.37%．

Abstract: In order to overcome the limitation of the existing alignment measures which need to set threshold for non-classification case-based reasoning systems,rank correlation is introduced to determine whether a case met CBR(case-based reasoning) hypothesis,in terms of which a few evaluation indicators for non-classification case-based reasoning systems are presented.The experiment is performed by testing 55 air investigation reports from Canadian Transportation Safety Board. The result shows that the rank correlation alignment method an acquire higher accuracy(10.91%and 16.37%,than case alignment and weighted correlation method respectively) in non-classification case retrieval.