Answers are grounded in this paper and cite their sources.
The study utilized 401 data points categorized into upper, middle, and lower levels. The model development process included five stages: 1) data purification and preprocessing, 2) embedding, 3) data segmentation and shape conversion, 4) model training, and 5) model performance evaluation. The research employed BERT, a pre-trained model, to develop the grading model. Performance evaluation of the trained model yielded an accuracy of 0.7377, precision of 0.6514, recall of 0.7377, and an F1 score of 0.6717. These results demonstrate a relatively high level of performance compared to previous studies on scoring Korean written and essay answers by L1 learners. As a result, the feasibility of developing an automatic scoring model using the BERT language model with small-scale data from a specific domain. Also, the model’s performance shows some generalizability, but further exploration is needed for improvement. Limitations of the study include the use of writing samples graded without considering grade levels, limited test data, difficulties in processing unregistered tokens, and the inherent unexplainability of deep learning techniques. Further discussions and considerations on how to more effectively utilize artificial intelligence in Korean language education should continue. Ongoing research on automatic grading is necessary to provide accurate and detailed educational feedback to students. As automatic grading research in the field of Korean language education advances, it is expected that high-quality educational interventions for students will become possible in the future.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: