Research on data modeling and strategy optimization algorithm of teacher guidance behavior for precision teaching

This study focuses on the data modeling and strategy optimization algorithm of teacher guidance behavior for precision teaching. In view of the lag and blindness of the traditional experience-driven guidance mode, this paper puts forward a trinity analysis framework of teacher guidance behavior, which is data-driven, dynamic modeling and intelligent optimization. In terms of modeling, a dynamic feature extraction method based on a multimodal spatiotemporal graph neural network (MST-GNN) is adopted. This method integrates classroom environments, teaching events, and multi-source sensing data to construct a graph structure with spatiotemporal characteristics, achieving behavior analysis from "event-level" to "interaction flow-level". For strategy optimization algorithm design, a dual-layer reinforcement learning optimization algorithm (DRL-Tutor) is proposed. It combines offline policy exploration with online fine-tuning. The offline policy exploration layer mines high-reward behavioral patterns from historical data to form macro strategies; the online fine-tuning layer dynamically adjusts strategies according to real-time classroom feedback. The experimental results of smart classroom platform show that MST-GNN is significantly superior to traditional methods in feature extraction, and DRL-Tutor is excellent in the prediction accuracy of time series behavior and the optimization of teaching strategies, which effectively improves students' knowledge retention rate and participation, shortens the time to correct cognitive bias, and has a low response delay in strategy generation, which provides a strong methodological support for the digital transformation of teachers' roles in the era of intelligent education and promotes the evolution of teaching decisions from experience to data science.