属性层级模型的改良及诊断性能研究  

Improvement of Classification Methodsand Accuracy of AHM

作　　者：韩裕娜[1,2] 张敏强[1] 方杰[3] 

机构地区：[1]华南师范大学心理应用研究中心,广州510631 [2]华南师范大学基础教育培训与研究院,广州510631 [3]广东财经大学人文与传播学院,广州510320

出　　处：《心理学探新》2015年第1期90-95,共6页Psychological Exploration

基　　金：全国教育科学“十二五”规划教育部重点课题(GFA111009);广东省哲学社会科学“十二五”规划2013年度一般项目(GD13CXL01);华南师范大学青年教师科研培育基金项目

摘　　要：该研究先简单介绍属性层级模型及其诊断步骤,然后分析方法 A和方法 B的计算公式及局限,对其进行改良得到两种新分类方法——方法 C和方法 D,并通过两个蒙特卡洛模拟实验来比较A、B、C、D四种方法的诊断性能,以及考察测验因素对四种分类方法诊断性能的影响。结果表明:(1)改良的方法比对应的原方法有更好的诊断性能。(2)各种方法诊断指标的高低跟属性结构有关。对线型结构进行诊断,D方法最优;对其他三类结构进行诊断,C方法最优。(3)各种方法的各种诊断性能指标随着失误水平参数变大而变小。Attribute Hierarchy Method（ AHM ）is one of the most influential cognitive diagnosis models, which was brought out by Leighton in 2004. Two traditional methods, Method A and Method B, based on item response theory are presented by Leighton to illustrate the classification of observed response patterns in the AHM. But their diagnostic aecuracy is not high. This study presents two new elassification methods, Method C and Method D, which are varied from the two traditional methods. Two Monte Carlo simulation experiments is designed to explore the performance of diagnosis on different condition. The results show： （ 1 ） the diagnostic performance of new methods is better than the traditional methods. The pattern match ratio（PMR） and marginal match ratio（MMR） of Method C is higher than Method A. The PMR and MMR of Method D is higher than Method B. （2） Performance of different classification methods is influenced by cognitive structure of items and number of items. On Linear structure, the performance of Method D is best. On other structures, the performance of Method C is best. （3）ff the slippage probability decreases, the accuracy of AHM will be improved.

关 键 词：认知诊断 AHM模型 判别方法 

分 类 号：B841.2[哲学宗教—基础心理学]