| 杨阳蕊,董方宁,王鹏斐,菅朋朋,李海昆.基于大语言模型的水利工程运行管理质量概念模型及知识图谱自动化构建[J].水利学报,2025,56(5):634-645 |
| 基于大语言模型的水利工程运行管理质量概念模型及知识图谱自动化构建 |
| Automated construction of schema and knowledge graphs for the operation and management quality of hydraulic projects based on large language models |
| 投稿时间：2025-04-09 |
| DOI：10.13243/j.cnki.slxb.20250027 |
| 中文关键词: 大语言模型 概念模型 知识图谱 智能生成 水利工程运行与质量管理 |
| 英文关键词: Large Language Models schema knowledge graph intelligent generation operation and management quality of hydraulic projects |
| 基金项目:国家自然科学基金项目（72271091）；河南省高等学校重点科研项目（25A520006）；华北水利水电大学硕士研究生创新能力提升工程项目（NCWUYC-202416098）；河南省科技厅科技攻关项目（252102210030） |
|
| 摘要点击次数: 214 |
| 全文下载次数: 135 |
| 中文摘要: |
| 现阶段水利工程运行管理质量相关数据大多存储在非结构化文本中，数字化程度较低，难以满足水利高质量发展提出的更高要求。现行知识图谱及概念模型构建方式严重依赖人工，效率欠佳。为此，本文提出一种基于大语言模型（LLMs）的“发掘-构建-过滤”（ECF）框架，以实现水利工程运行管理质量概念模型及知识图谱的自动化构建。该框架利用LLMs首先发掘出知识图谱的实体和关系类型，设计并生成知识图谱概念模型。随后，在该概念模型的指导下，从数据源中提取实例，构建知识图谱。最后，设计过滤机制，剔除概念模型及知识图谱中的三元组噪声，保证准确性。通过设置种子文本集、全体文本集数据，对ECF框架各环节进行评估并与现有方法进行对比。结果表明，ECF框架在概念模型及知识图谱的自动化构建方面表现良好，三元组准确率较现有方法提升23%，优化了知识图谱的构建效率，为水利工程的规范运行与稳步推进提供了技术和理论支持。 |
| 英文摘要: |
| At present，the quality management related data of hydraulic projects are mostly stored in unstructured text with a low degree of digitization，making it difficult to meet the higher requirements for high-quality development. To overcome the shortcomings of the current knowledge graph and knowledge graph schema construction methods，which rely heavily on manual labor and have poor efficiency. This paper proposes an Explore-Construct-Filter （ECF）framework based on large language models（LLMs）to achieve automated construction of conceptual models and knowledge graphs for the quality management of hydraulic project operation. The framework uses LLMs to first discover the entities and relationship types of the knowledge graph，and then designs and generates a conceptual model of the knowledge graph. Subsequently，under the guidance of the conceptual model，instances are extracted from the data source to construct a knowledge graph. Finally，design a filtering mechanism to remove triplet noise from conceptual models and knowledge graphs，ensuring accuracy. By setting the seed text set and the entire text set data，the various components of the ECF framework are evaluated and compared with the existing methods. The results show that the ECF framework performs well in the automatic construction of conceptual models and knowledge graphs，with an accuracy rate 23% higher than that of existing methods，thus optimizing the efficiency of knowledge graph construction，and providing technical and theoretical support for the standardized operation and steady progress of water conservancy engineering. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |