AI-Supported Evaluation of Reflective Learning in Creative Practice Courses

Abstract

This study examines how AI-supported text analysis can be used to evaluate reflective learning in creative practice courses. The study is designed to collect approximately 2,400 reflective journals, 1,200 learning logs, 600 project statements, and 480 teacher feedback records from 400 students enrolled in creative practice courses over a 16-week semester. The research measures reflective depth, conceptual development, emotional expression, self-assessment quality, problem identification, revision awareness, and learning strategy use. Natural language processing methods, including topic modeling, sentiment analysis, semantic similarity analysis, text classification, and large language model-assisted rubric scoring, are used to analyze students’ written reflections. The study further compares AI-generated scores with teacher-rated scores through correlation analysis, inter-rater reliability testing, mean absolute error, and intraclass correlation coefficient. Regression analysis is used to test whether reflective depth and revision awareness significantly predict project performance and learning improvement. The innovation of this study is that it does not treat reflective writing as purely qualitative evidence, but converts reflective learning into measurable textual indicators while preserving teacher interpretation as the final evaluative reference.