A test theory based on deep learning: Deep Response Model

Abstract

In this article, Deep Response Model(DRM), a novel deep neural networks model for estimating student characteristic parameters and predicting student's response patterns is proposed. It consists of two independent neural networks, Student Layer and Item Layer, combines their outputs and predict student's response. Then, values of the last student layer are considered as student parameters. It can estimate student parameters and predict their responses with high accuracy even when it is difficult to calibrate item parameters.