A Complex Reasoning Framework for Intelligent Learning Assistant Based on Large Language Model and Knowledge Graph

Large language models are widely applied in educational scenarios. Students can engage in personalized and interactive learning through intelligent learning assistants installed on various devices. Among these applications, integrating large language models (LLMs) with knowledge graphs to achieve complex reasoning is a key issue. Traditional methods often solve multi-hop query problems based on individual technologies, leading to insufficient accuracy and comprehensiveness in answering complex questions. This paper proposes an intelligent agent framework called K12-Agent, which uses retrieval-enhanced generation, fine-tuning of LLMs integrated with knowledge graphs, and collaborative reasoning to enhance the initiative of the agent in interactions with students. Compared to traditional methods, K12-Agent has improved the accuracy by 15% and the recall rate by 13% when facing complex student questions. Additionally, K12-Agent is more adaptable to the expression methods of younger students, making it more user-friendly.