Automatic Scoring Method of Short-Answer Questions in the Context of Low-Resource Corpora

Short-answer questions offer a method to evaluate whether students have acquired a domain of knowledge. Because it is very time-consuming to evaluate the answer of the questions, many automatic scoring methods have been proposed. However, most of these models require large quantities of training data. This study proposes a method for automatic creation of a concept map based on small quantities of corpora. The concept map verified by experts is available and reliable. In addition, this study proposes a method to score the correctness of students' answers to short-answer questions using a machine-generated concept map. The preliminary experiment shows that the correctness of scoring results of the proposed method is close to that of the BERT-based model. Since the proposed method does not require large quantities of training data, it is suitable for the instruction system only using low-resource corpora.