Automated Machine Learning for Short-essay Answer Grading

Abstract

Natural language processing coupled with machine learning has been employed to great effect in the realm of essay and short-answer essay scoring. However, these models typically require a strong understanding of machine learning concepts along with domain knowledge specific to the dataset in order to achieve optimal results. An emerging trend in machine learning is that of automated machine learning. Our project seeks to demonstrate the effectiveness of an automated machine learning system designed to score a critical thinking assessment with no machine learning experience or domain knowledge required. We will show that an ensemble of deep and non-deep machine learners, designed and tuned automatically, can outperform human scorers on this task.