基于属性层级关系的rRUM模型优化——模型解释力及判准率的提升视角

以提高认知诊断模型判准率及对数据的解释力为视角,对当前应用较广泛的rRUM模型进行优化(优化后的模型简记为rRUM-AH),并采用Monte Carlo模拟研究及实证研究相结合的范式,比较分析了传统的rRUM模型和rRUM-AH模型的诊断正确率及诊断结果的解释力.研究结果表明:当属性间存在层级关系时,不论在何种实验设计条件下,优化后的rRUM-AH模型属性诊断正确率远远高于传统的rRUM模型; 当属性间存在层级关系时,rRUM模型的模式判准率平均不到80%(而rRUM-AH模型平均高达90%以上),难于满足实际需求,此时实际应用者选用该研究新开发的模型是一个较好的选择.

The Revision of the Reduced RUM Based on the Attribute Hierarchy

The Monte Carlo simulation method and empirical study were both used here to investigate the accuracy and the power of output explanation of diagnosis.Three simulation studies were conducted:Study 1 was based on the fix sample size,test length and the number of attributes.While Study 2 and Study 3 were based on the variable sample size and test length and the number of attributes respectively.Different attribute hierarchies were contained in all three simulation studies.At last the real data about syllogistic reasoning was chosen as an example to illustrate the application and comparison of the rRUM and the rRUM-AH.The findings showed that:If the relationship between attributes satisfied some hierarchy structure,then under any experiment design condition,the corrected ratios under rRUM-AH model were greater than those under rRUM model.When the relationship between attributes satisfied some hierarchy structure,the average of pattern corrected match ratios under rRUM model was less than 80 percent,otherwise,it under rRUM-AH model was greater than 90 percent.To satisfy the need in practice,the rRUM-AH model is a good choice for practitioner.Based on the analysis on the syllogistic reasoning data,the outputs indicated that the rRUM-AH model could fit the data better than rRUM model,and the power of result explanation under it was more reasonable.