Diffusion Model-Assisted Interactive Scene Generation and Student Participation Prediction for Ideological and Political Education in Textile Majors

To address the insufficient integration between textile engineering knowledge and ideological–political education, as well as the lack of real-time assessment of student engagement, this study proposes a diffusion model-assisted framework for interactive scene generation and multimodal participation prediction. A dual-encoder diffusion architecture driven by a Textile Knowledge Graph (TKG) is developed to establish semantic associations between textile process parameters and ethical objectives, enabling the automatic generation of process-aware educational scenarios. Eye-tracking and electromyography signals are jointly modeled using a Long Short-Term Memory network to predict participation status, while a dynamic feedback mechanism adaptively adjusts scene complexity according to learners’ cognitive responses. Considering the growing adoption of intelligent sensing and distributed information acquisition technologies in digital education environments, the proposed framework provides a scalable solution for multimodal data fusion and adaptive interaction, offering potential methodological references for future wireless sensing and smart educational infrastructures. Experimental results demonstrate that the proposed method achieves a process parameter trigger rate of 82.1%, an ethical scene coverage of 93.2%, classification accuracies exceeding 85% for highly engaged students, and interaction latency below 50 ms under low-complexity conditions. The framework effectively enhances scene authenticity, participation prediction accuracy, and adaptive instructional capability, providing a reliable technical foundation for intelligent textile education and interdisciplinary human–machine interaction systems.