Verification of Deep Gaussian Process Regression for Automated Essay Scoring

Abstract

Automated essay scoring of descriptive answer data is known to achieve high accuracy in score prediction by using regression models based on language models.In the operation of test services, there is often a need to determine the uncertainty of predicted values, which has been challenging with conventional methods and can lead to decreased prediction accuracy in classification models.In this study, we investigate regression models that can improve prediction accuracy while also calculating uncertainty compared to conventional methods. Specifically, we verified Gaussian process regression models and deep Gaussian process regression models using both benchmark datasets and our own large-scale dataset for accuracy validation. The experimental results demonstrate the effectiveness of the deep Gaussian process regression model.