一种新的基于神经网络的IRT项目参数估计模型

探讨了一种新的基于广义回归神经网络（GRNN）的IRT（项目反应理论）项目参数估计建模方法，着重介绍了如何建立网络的输出模式及利用Monte Carlo方法建立网络的输入模式，提出了多种对模型进行改进的方法。模拟实验表明，利用GRNN可以以任意精度拟合CTT（经典测验理论）参数统计值和IRT参数值间隐含的非线性关系。与其他方法进行的比较表明，在小样本情况下，该方法的参数估计误差更小。

New model of IRT item parameter estimation based on neural networks

A new modeling method based on general regression neural networks(GRNN) of item parameter estimation within IRT(Item Response Theory) was discussed. The methods about how to construct the output pattern of neural networks, and especially the input pattern by using Monte Carlo method were described. Methods about how to improve the learning efficiency and generalization ability have been proposed. Simulation experiments denote that it is feasible to fit the nonlinear relationship of item parameters between CTT(Classical Test Theory) and IRT given any precision. Comparisons of this method to other methods have been done at last, which suggested it somewhat advantageous.