大模型提取SQL字段血缘的方式研究  

Research on SQL Field Lineage Extraction Method Based on Large Model

作　　者：梁思捷 梁丽霞 王树发 陈星 陈声晓 LIANG Sijie;LIANG Lixia;WANG Shufa;CHEN Xing;CHEN Shengxiao(Department of Information,Guangdong Tobacco Guangzhou Co.,Ltd.,Guangzhou Guangdong 510610,China;Department of Technology,Yueshutech,Guangzhou Guangdong 510610,China)

机构地区：[1]广东烟草广州市有限公司信息中心,广东广州510610 [2]广州阅数科技有限公司技术部,广东广州510000

出　　处：《信息与电脑》2024年第10期84-87,共4页Information & Computer

基　　金：穗烟科项202401。

摘　　要：本文研究了一种基于大模型的结构化查询语言(Structured Query Language,SQL)字段血缘提取方法,利用大模型的强大的语义表示能力,从SQL语句中提取字段和表之间的数据流向和变化规则,从而实现数据治理和分析的基础。本研究基于ChatGLM 3.0开源大模型预参数和LoRA(Low-Rank Adaptation)微调模型,通过实验对比了四个商用大型语言(Large Language Model,LLM)模型的案例分析证明了方案的可行性。This paper studies a SQL field lineage extraction method based on large model,which uses the powerful semantic representation ability of small-parameter large model to extract the data flow and change rules between fields and tables from SQL statements,thus providing the basis for data governance and analysis.The main method is based on the pre-parameters of ChatGLM 3.0 open source large model and LoRA fine-tuning model,and the feasibility of the scheme is proved by the case analysis of four commercial LLM models.

关 键 词：大模型 结构化查询语言 字段血缘 Lora微调 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]