A BERT and Text Graph Neural Network-Based Fusion Model for Automated Short Answer Grading

Automated Short Answer Grading (ASAG) is a popular task in education technology that automates the assessment of student-written short answers by leveraging machine learning and Natural Language Processing (NLP) techniques. In recent times, the use of Graph Neural Networks (GNNs) with appropriate graph representations of the textual data has shown promising results in different NLP tasks. This paper proposes a fusion model for the ASAG task that integrates the advantages of semantic and contextual embedding of the transformer-based BERT model with the benefits of GNN-based textual features representing document-level structural information. An empirical study is performed over the publicly available ASAP dataset to test the performance of the proposed fusion method. It is observed that our proposed model outperformed the other state-of-the-art methods by securing a 0.885 F1 score in the four-class classification task.