Validation of a Japanese essay proofreading method based on sentence similarity

Abstract

In recent years, machine learning technology has made great progress, and machine learning is now used for various tasks in the field of natural language processing. Among them, there is research on automatic grading of essay-type questions, and various methods have been proposed. It is considered difficult to automatically evaluate essay-type questions for which there are no clear-cut correct answers. The purpose of this study is to construct a Japanese essay proofreading assistance system that can predict the factors necessary for automatic scoring and improvement of evaluation. We verified the performance of the deep language models, BERT and Sentence-BERT, on this system.