Exploring large language model applications in dance education from an educational psychology perspective

Area of Science:

• Artificial Intelligence
• Educational Technology
• Dance Science

Background:

• Traditional dance instruction often lacks personalization and fails to adapt to individual cognitive and emotional needs.
• Large Language Models (LLMs) offer potential for creating adaptive and personalized learning experiences.

Purpose of the Study:

• To propose a psychology-informed framework for integrating LLMs into dance education.
• To develop a conceptual methodology, DanceGPT, for personalized dance instruction and performance evaluation.

Main Methods:

• Designed DanceGPT as a multimodal architecture integrating visual, motion, and textual data.
• Implemented a core multimodal video-based action quality assessment model using computer vision and pose estimation.
• Utilized multimodal fusion, domain-guided attention, and learner-aware conditioning for performance evaluation.

Main Results:

• The proposed model demonstrated consistent improvements in ranking correlation and regression accuracy compared to baselines.
• Effectiveness of multimodal fusion and adaptive knowledge integration for dance performance evaluation was shown.
• Educational psychology principles were used for personalization and adaptive feedback design.

Conclusions:

• The framework shows promise for enhancing dance education through AI-driven personalization and feedback.
• Further validation through real learner studies and interactive tutoring is recommended.
• Future research should focus on longitudinal evaluations in authentic dance settings.