Automation of Short Answer Grading Techniques: Comparative Study using Deep Learning Techniques

Abstract

Automatic short answer grading (ASAG) techniques have been shown to cut down on the time and work needed to grade exams, and it is a method that is becoming more and more common, especially with the rise in popularity of online courses. This study compares the results of 7 pre-trained embedding models using just one feature to automatically grade brief responses: the similarity between the model answer's and the student answer's embeddings. Regression models are developed and evaluated to predict a short answer's score based on the similarities between all pairs of answers in the Mohler dataset. The predictions are evaluated by comparing the Root Mean Squared Error (RMSE) and Pearson correlation scores of each model.