摘要
计算机化自适应测验(CAT)的选题策略是影响测量准确性的主要因素之一.针对认知诊断CAT初期知识状态估计不准确的问题,改进后验加权Kullback-Leibler信息量,得到了2种平均后验加权Kullback-Leibler信息量选题指标.然后运用DINA(the deterministic inputs,noisy"and"gate)模型模拟作答反应,在不同测验长度下比较了6种选题策略的优劣.结果表明,新指标能极大地提高测量准确度,当测验长度为15时,知识状态的判准率提高了10%以上;除随机方法外,其他方法的项目曝光率没有明显差异.
The item selection method in cognitive diagnostic computerized adaptive testing (CD-CAT) is one of the main factors influencing measurement accuracy. In view Of inaccurate estimated knowledge state in the early stage of CD-CAT, two forms of averaged posterior weighted Kullback-Leibler information were derived from posterior weighted Kullback-Leibler information. Results of six item selection methods were compared under different test length, using the deterministic inputs, noisy "and" gate (DINA) model to simulate the response of students. Data showed that the new item selection methods improved CD-CAT measurement accuracy. With the test length at 15, classification accuracy of knowledge state increased by more than 10%. Other than random selection method, item exposure rates of other methods had no significant difference from each other.
出处
《北京师范大学学报(自然科学版)》
CAS
CSCD
北大核心
2011年第3期326-330,共5页
Journal of Beijing Normal University(Natural Science)
基金
教育部新世纪优秀人才支持计划资助项目(NCET-07-0097)
关键词
认知诊断
计算机化自适应测验
选题策略
测量准确性
cognitive diagnose
computerized adaptive testing
item selection method
measurement accuracy