摘要
针对标准参照测验题库建设中存在的问题,提出了运用广义回归神经网络集成来估计标准参照测验的IRT项目参数的新方法,讨论了建立神经网络集成的理论依据,给出了实现方法,并以单参数的Logistic模型为研究对象进行模拟实验研究.结果表明,在较小考生样本情况下,相对于传统IRT方法,神经网络集成可以得到远远优于它的结果.
Aiming at the problem of item pool construction in criterion-referenced test, a new method applying general regression neural network ensemble to estimate IRT item parameter in criterion-referenced test is proposed. The elementary principle that why neural network ensemble can be constructed is demonstrated, as well as and the method about how to construct the network. And simulated experiments are conducted with the research object of one parameter logistic model. The result shows that under the condition of small size of examinees, it is far better than that drawn from conventional IRT parameter-estimation methods.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2006年第B07期36-39,共4页
Journal of Harbin Engineering University
关键词
神经网络集成
项目反应理论
小样本
参数估计
neural network ensemble
item response theory
small samples
parameter-estimation
作者简介
余嘉元(1949-),男,教授,博士生导师
汪存友(1982-),男,硕士研究生