Confucius: A Knowledge Graph-Enhanced LLM Agent for Intelligent Teaching Video Supervision

With the deepening of educational reform and the widespread adoption of online learning, classroom video analysis has become crucial for improving teaching quality. However, significant challenges remain in multimodal data fusion and semantic understanding. To address this, we propose an off-campus educational monitoring platform based on large language models (LLMs) and knowledge graphs. By incorporating structured educational knowledge, our platform effectively mitigates LLM hallucinations while enhancing accuracy in instructional context analysis and sensitive content detection. Experimental results demonstrate its strong performance in violation identification and scenario comprehension. Pilot applications further validate its practical value in improving education supervision efficiency, offering a viable solution for intelligent education management.