Large Language Model-driven knowledge graph construction for Qing Yongzheng imperial artifact production

This study constructs a knowledge graph for Qing imperial artifacts during the Yongzheng period, utilizing the Huoji Dang archives to systematize fragmented historical records and establish a robust methodological framework for Qing material culture research. Employing Large Language Models (LLMs) integrated with a Retrieval-Augmented Generation (RAG) framework and carefully optimized prompts, we successfully automate the extraction of unstructured data and store the resulting structured entities within Neo4j. Experimental results demonstrate DeepSeek-V3’s superior performance under RAG augmentation, achieving a Micro F1 score of 86.75% (Precision:80.54%/Recall:94.00%) and a Macro F1 score of 84.70% (Precision:83.29%/Recall:88.39%). The generated knowledge graph visually reveals the underlying organizational hierarchies, intricate production workflows, and prevailing imperial esthetic preferences embedded within these archival records. The primary contribution of this work lies in enabling the systematic structuring of historical texts via graph technology, thus providing a unified platform for the study of Qing imperial material culture. This approach significantly enhances traditional research paradigms by integrating advanced artificial intelligence with digital humanities methodologies.