Employing SBERT for Essay Assessment

Conducting exams is important not only to evaluate the performance of students but also to analyze that of learning process. However, it is not easy and even time consuming to assess exams, especially for essays. To address such issues, it is worthwhile to have a system that can help in assessing exams. In this study, we aim to develop an automated essay scoring system to assess essay exams in Cibinong Mechanical Vocational High School, Bogor, Indonesia. We utilize semantic similarity evaluation and syntactic analysis to enhance the evaluation system of the Cibinong Mechanical Vocational High School for essay answers. Specifically, we employ Sentence Bidirectional Encoder Representation from Transformer (SBERT) as our model. Our SBERT model results show that they could perform an automatic assessment as good as the conventional one in evaluating the essay exams. In particular, the average mean absolute error (MAE) between human assessment and SBERT one was 0.26. Then, the SBERT model is adopted to develop the automated essay scoring system to help Cibinong Mechanical Vocational High School staff assess their students’ exams.