Abstract: This study conducts intelligent semantic knowledge organization and questionanswering application research on the special literature Produce in Local Chronicles based on Large Language Models (LLMs) and Graph RetrievalAugmented Generation (GraphRAG) technology. By constructing a knowledge graph for produce in local chronicles and implementing an intelligent questionanswering system, the accessibility and practical value of ancient Chinese texts are effectively enhanced, providing a new paradigm for inheriting and promoting local culture. Taking data from the Yunnan Volume as an empirical object, the research designed structured prompts using the CRISPE framework and evaluated ten mainstream large models such as GPT4o and DeepSeekV3 through multidimensional indicators. Experimental results show that the QwenLong model demonstrates the optimal performance in entityrelation extraction. The knowledge graph constructed by QwenLong successfully integrated 24,215 triples, realizing semantic correlation of previously scattered produce knowledge. Building on this graph with GraphRAG and LangChain framework, an intelligent questionanswering system was built. The knowledge graphenhanced DeepSeekV3 model achieve the best performance across six categories of representative question with an accuracy rate of 86.33%, capable of providing accurate and coherent interactive knowledge services. This research validates the feasibility of LLMbased technology to knowledge organization of ancient Chinese texts and offers methodological references for deep integration of digital humanities. 

Key words: Large Language Models (LLMs), Produce in Local Chronicles, Knowledge graph, Intelligent questionanswering, Graph RetrievalAugmented Generation (GraphRAG)