山西大学学报（自然科学版）2025，Vol.48Issue(5)：859-870,12.DOI:10.13451/j.sxu.ns.2025012
基于LoRA微调的撒拉族建筑知识图谱构建
Construction of Knowledge Graph for Salar Architecture Based on LoRA Fine-tuning
摘要
Abstract
In the context of promoting the excellent traditional Chinese culture,the unique Salar culture in the Methodist region of Qinghai Province has become an important part of the tourism resources in Qinghai Province.Salar architecture covers the Salar hu-man history,politics and religious culture,but at present the construction technology of Salar architecture is gradually lost,and the inheritance and protection of Salar architecture is urgent.As a mainstream knowledge digitization technology,the combination of knowledge graph and Salar architecture can better protect and inherit the Salar architecture.Therefore,this paper constructs a unique Salar architecture dataset and ontology structure,uses the large language model LoRA fine-tuning method to extract entities and rela-tions from Salar architecture data,and conducts comparative experiments with the classical knowledge extraction model.Compared with ChatGPT,the F1 values of entity extraction and relation extraction are improved by 13.37％and 16.32％.The knowledge graph constructed by the institute can be used for knowledge recommendation,intelligent search and knowledge graph enhanced question answering system based on RAG technology.关键词
大语言模型/LoRA微调/知识抽取Key words
large language model/LoRA fine-tuning/knowledge extraction分类
数理科学引用本文复制引用
马咏梅,耿生玲,赵维纳,马龙龙,安波..基于LoRA微调的撒拉族建筑知识图谱构建[J].山西大学学报（自然科学版）,2025,48(5):859-870,12.基金项目
国家自然科学基金(61862055 （）
62266036) （）
国家社会科学基金(22BTQ010) （22BTQ010）
科技援青合作专项-区域处(2022-QY-203) （2022-QY-203）
中国社会科学院专项(2024SJK017) （2024SJK017）
