摘要
为探讨海明距离判别法(HDD)的非参数优势,通过一个5因素混合实验,考察了4个因素(属性层级、测验长度、样本容量、知识状态分布)对HDD的3种判别方法(R方法、B方法、W方法)分类准确率的影响.结果表明:1)属性层级和测验长度均会影响HDD判准率,属性层级越紧密、测验长度越长,HDD判准率越高;2)HDD对样本容量无依赖,可适于小样本评估;3)HDD的R方法、B方法、W方法的分类准确率无差异;4)HDD无需被试知识状态分布的正态性假设,更适于均匀分布.
In order to explore the advantages of non-parametric methods in cognitive diagnostic assessment and better understand the properties of HDD,Monte Carlo simulation method is used to examine a variety of factors that influence diagnostic accuracy. Fixed the test measured attributes as five,a mixed-factor design is conducted to investigate the effect of attribute hierarchy,test length,sample size and knowledge state distribution on HDD three different discriminating methods( i. e.,Random method,Bayes method and Weighted method) classification accuracy. And the results showe that both attribute hierarchy and test length have an impact on the classification accuracy of HDD method,the closer of the test's attributes relate and the longer of test length,the higher of the HDD method classification accuracy. And the sample size has little effect on HDD,so it is suitable for small scale implementation. There is no difference between three HDD methods of classification accuracy. Moreover,the normal distribution of knowledge state is not necessary when applying HDD to test Data,it performs a little better when the knowledge state is uniform.
出处
《江西师范大学学报（自然科学版）》
CAS
北大核心
2017年第4期394-400,共7页
Journal of Jiangxi Normal University(Natural Science Edition)
基金
教育部人文社会科学研究一般课题(16YJA190002)资助项目
关键词
海明距离判别法
属性层级
测验长度
样本容量
知识状态分布
Hamming distance discrimination method
attribute hierarchy
test length
sample size
knowledge state distribution
