基于神经网络的项目参数估计方法

Method of Item Parameter Estimation Based on Neural Networks

针对题库建设中项目参数估计的实际问题,提出了一种全新的基于神经网络的参数估计方法;并以二值记分的3PLM为项目反应理论模型,以广义回归神经网络为网络模型,根据Monte Carlo实验法进行了模拟实验研究,最后将神经网络方法与传统的数理统计估计方法进行了比较。结果表明,在小样本测验情况下,神经网络方法具有一定的优势,尤其是当去掉对项目参数的先验概率分布的限制时,神经网络方法的优势更加明显,说明本文提出的方法具有一定的价值。

To solve the problem on item parameter estimation when constructing item bank, a new method based on neural networks （NN） is proposed here. And by choosing three-parameter logistic model （3PLM）, general regression neural networks （GRNN） and Monte Carlo as the item response theory （IRT） model, NN model and experimental method respectively, some simulation experiments are conducted. To inspect whether our method works well, a comparison is carried out between the NN method and traditional method, which indicates that the former is more advanta-geous, especially when the traditional method cancels its item prior constrains.