认知诊断计算机化自适应测验的题库使用均匀性初探

项目曝光控制是认知诊断计算机化自适应测验(CD-CAT)中亟需解决的重要问题之一。采用蒙特卡洛模拟方法对CD-CAT中五种常用选题策略(随机化方法、KL信息量方法、香农熵方法、后验加权的KL信息量方法和综合后验加权和距离加权的KL信息量方法)的题库使用情况进行探讨。结果发现:四种非随机化选题策略的题库使用均匀性较差、测验重叠率高,从而导致测验安全性较差;香农熵方法的判准率总是最高。今后可以将传统CAT中的项目曝光控制技术融入到CD-CAT选题策略中。

A NOTE ON THE UNIFORMITY OF ITEM BANK USAGE IN COGNITIVE DIAGNOSTIC COMPUTERIZED ADAPTIVE TESTING

Item exposure control is one of the important issues which should be solved in cognitive diagnostic computerized adaptive testing(CD-CAT).Monte Carlo simulation method was used to discuss the uniformity of item pool usage resulted from five common item selection strategies(random item selection strategy,KL information method,Shannon Entropy method,Posterior Weighted KL method and Hybrid KL method)in CD-CAT.The results indicated that:in the four non-randomized item selection strategies,the uniformity of item pool usage is poor,and the test overlap rate are considerably high,which make the test security poor.Shannon Entropy method always produce the highest pattern classification correct rate.Some item exposure control techniques commonly used in regular CAT(such as the a stratification method and the maximum priority index method)can be incorporated into the item selection strategy in CD-CAT in the future.