An Automatic System for Essay Questions Scoring based on LSTM and Word Embedding

Abstract

Essay scoring is a significant task in education, especially online education. Manual essay scoring is a complex job, which limits the development of large-scale online education. In tradition, essays are all scored by human because computer program cannot understand what text means. Recent advance of natural language processing technology in artificial intelligence provides a way to score essays automatically. In our study, we create an automatic essay scoring (AES) system by using Long-Short Term Memory (LSTM) network and word embedding. We show an automated system that can rate essays in electronic text. We combine manually crafted features and Word2Vec embedding in training the model, which makes it more interpretable. We carefully tune the hyperparameter to improve the precision of our model. The LSTM network reaches a quadratic weighted Kappa score (QWK) of 0.95 ± 0.01, which outperforms many other rating systems. The AES system we design can greatly improve the efficiency of online education.