基于大语言模型的档案文献深层语义内容结构生成式框架构造——以延安红色档案文献为例  

Research on Deep Semantic Content Structure Generation of Archival Documents Based on Large Language Models:A Case Study of Yan'an Red Archival Documents

作　　者：周文杰 赵悦言 朱召师 ZHOU Wenjie;ZHAO Yueyan;ZHU Zhaoshi(School of Information Resource Management,Renmin University of China,Beijing 100782,China;School of Public Administration,Sichuan University,Chengdu 610065,China;Central Archives,Department of Policy and Regulations,Beijing 100095,China)

机构地区：[1]中国人民大学信息资源管理学院,北京100872 [2]四川大学公共管理学院,成都610065 [3]中央档案馆政策法规司,北京100095

出　　处：《档案学通讯》2025年第2期13-21,共9页Archives Science Bulletin

基　　金：国家自然科学基金面上项目“循证信息贫困研究”(72374170)。

摘　　要：大语言模型的快速发展给档案管理的智能化转型提供了创新方向。以延安红色档案文献为例,构建基于LinkML语言描述的档案文本深度语义结构,利用SPIRES算法,以生成式逻辑向大语言模型(GPT 3.5turbo 16k)发出提示词,递归提取档案语料中包含的深层语义内容,并将其输出为结构化档案数据。综合效果评估显示,得益于大语言模型的强大文本内容解析能力,该方案无须依赖新的训练数据即可执行档案文献内容解析与表征任务,能够初步实现零样本条件下档案资源的精细化处理和开发,从而为大语言模型赋能档案信息化建设和智能化转型提供了强有力的技术支撑和创新思路。同时,大语言模型在档案文献语义解析的精确性等方面还有进一步优化的空间。The rapid development of large language models(LLMs)has provided innovative directions for the intelligent transformation of archival management.Using Yan'an Red Archives as an example,we constructed a deep semantic structure of archival texts based on LinkML language descriptions.Utilizing the SPIRES algorithm,we posed questions to the generative logic of the large language model(GPT-3.5 Turbo 16k)to recursively extract deep semantic content embedded within the archival corpus,and output it as structured archival data.Comprehensive performance evaluations indicate that this approach has significant advantages in accuracy and intelligence,achieving fine-grained processing and development of archival resources.It provides robust technical support and innovative ideas for empowering LLMs in the informatization and intelligent transformation of archival management.

关 键 词：大语言模型 档案资源开发 档案语义架构 结构化数据 

分 类 号：G273.5[文化科学—档案学]