A Question Answering Agent Integrating Knowledge Graph and Large Language Model

To address nges of insufficient kthe challenowledge accuracy, weak dynamic updating capabilities, and poor domain adaptability in vertical domain applications of large language models, this study proposes a question answering agent integrating knowledge graph and large language model. The framework optimizes entity labeling through Chain-of-Thought technology to construct fine-grained domain taxonomies and implements a “reasoning-action” collaboration mechanism using the React framework for multi-turn retrieval-verification cycles. Experimental results demonstrate that the framework significantly enhances the accuracy and information completeness of responses through synergistic optimization between structured knowledge and generative models, while exhibiting robust generalization capabilities and dynamic knowledge adaptability across diverse domain scenarios. The technical advantages of synergistic optimization between structured knowledge and generative models are validated through multi-domain evaluations.