The Application of Artificial Intelligence in Quality Control of Subjective Items Scoring in Large-scale Examinations

This paper focuses on the task of scoring quality control of subjective items in large-scale exams. We collected real student responses and human grading data from various subjects in the 2024 National College Entrance Examination in one of the provinces in China, and trained a scoring model. By comparing machine-generated scores with human scores and implementing a designed screening strategy, we ultimately identified samples that require re-evaluation by expert graders for final scoring. This method only requires a small amount of human scoring to train the model. Experiments on real data demonstrate that the proposed approach effectively identifies responses that need re-evaluation, thereby improving the overall grading quality.