Research on Teaching Video Monitoring Platform Based on Large Language Model Prompt Engineering

With the deepening of educational reforms and the rapid development of online education, the analysis of classroom teaching videos has gradually become an important means to enhance teaching quality and effectiveness. The information contained in teaching videos, such as teacher-student interactions and teaching behaviors, holds significant value for teaching quality assessment and strategy optimization. However, existing research often lacks effective integration capabilities when dealing with multimodal data, with insufficient semantic understanding and contextual awareness. Moreover, there is still room for improvement in the accuracy and interpretability of identifying inappropriate behaviors and assessing teaching quality. To address these issues, this study has developed a software system based on large language model prompt engineering and retrieval-augmented generation technology. By deeply analyzing text, audio, and image information in videos, combined with speech recognition and facial recognition technologies, the system can accurately analyze classroom teaching content and behaviors, providing objective and precise teaching feedback. The main contribution of this paper lies in proposing an analytical framework that significantly enhances the integration of multimodal data and semantic understanding capabilities, and in constructing a multimodal teaching video monitoring platform capable of automatically identifying inappropriate behaviors and providing interpretable analysis results. Experimental validation shows that the system excels in image-sensitive content recognition, contextual awareness, and video analysis, addressing the shortcomings of existing models and demonstrating significant practical application value.