Deep Item Response Theory as a Novel Test Theory Based on Deep Learning

Item Response Theory (IRT) evaluates, on the same scale, examinees who take different tests. It requires linkage of examinees’ ability scores as estimated from However, IRT techniques assume independently random sampling abilities a standard normal distribution. Because this assumption, not only much labor to design, but it also has no guarantee optimality. To resolve that shortcoming, study proposes novel based deep learning, Deep-IRT, which assumption randomly sampled Experiment results demonstrate Deep-IRT estimates more accurately than traditional does. Moreover, can express actual distributions flexibly, merely following distribution assumed for IRT. Furthermore, show predicts examinee responses unknown items examinee’s own past response histories