Multi-Agent Knowledge Graph Question Answering Framework Based on Lightweight Large Language Model

Industrial knowledge graphs pose significant challenges for question answering due to their logical complexity and obscure business knowledge, making it difficult to leverage existing technologies. This paper proposes a unified framework based on lightweight large language model agents to collaboratively process industrial knowledge graphs, enabling knowledge-questions-answering in complex scenarios and business logic. The lightweight large language model we released can accurately identify complex entities in industrial data and possesses excellent knowledge-enhanced reasoning capabilities, while meeting the low-resource and high-performance requirements of industrial knowledge graph question answering. Experimental results demonstrate that in a cold-start setting, our proposed framework achieves 73 percent accuracy on industrial knowledge graphs, with an average improvement of $\mathbf{4 6}$ percent.