基于IRT模型的BP神经网络参数估计的进一步研究

Further Research on the Parameter Estimation Method of BP Neural Network Based on Item Response Theory

对基于IRT模型的BP神经网络降维法参数估计中的BP神经网络的网络隐层数及隐层节点数进行改进,并对其降维法中不合理的部分予以修正.通过蒙特卡洛模拟研究结果表明:对参数估计的各评价指标,改进的方法均优于原方法.

Parameter estimation is one of the important components of item response theory（IRT）. When the sample of examinees or the number of the items is small,the traditional statistical methods of parameter estimation may be failure. Due to the shortcoming of the traditional parameter estimation methods,a new way based on BP neural net-work with reduced dimension method was proposed. This article is a further research on the work exited. With a modified reduced dimension method and a new construction of BP neural network,the Monte Carlo study shows a better result.