基于文本大模型的财务本地智库系统的设计与实现  

Financial local think-tank system based on large language model

作　　者：翟鲁杰 张鹏 于健 刘洪军 ZHAI Lujie;ZHANG Peng;YU Jian;LIU HongJun(Financial Shared Service Center,China Railway Jinan Group Co.Ltd.,Jinan 250001,China)

机构地区：[1]中国铁路济南局集团有限公司财务共享服务中心,济南250001

出　　处：《铁路计算机应用》2025年第4期82-86,共5页Railway Computer Application

摘　　要：针对财务共享服务中心及财务共享服务管理信息系统在推广应用过程中用户频繁培训及重复提问的问题,设计了一套基于文本大模型的财务本地智库系统(简称:本地智库系统)。通过内网本地部署国产开源文本大模型,结合LangChain框架,搭建了本地智库系统的整体架构,并详细阐述了该系统的功能及关键技术。该系统依托向量数据库中上传的文档资源,经过后期微调优化处理,提供了生成式对话问答服务,辅助财务共享服务中心工作人员开展线上和线下培训工作。实验验证表明,该系统能够准确回答用户提出的问题,减轻了培训和推广人员的工作负担,在铁路其他专业培训工作方面也具有应用前景。In response to the frequent training and repeated questioning of users during the promotion and application of the Financial Shared Service Center and financial shared service management information system,this paper designed a financial local think-tank system(referred to as the local think-tank system)based on a large language model,built the overall architecture of a local think-tank system by deploying a domestic open-source large language model locally on the intranet,combined with the LangChain framework,and elaborated on the system's functions and key technologies in detail.The system was relied on the document resources uploaded from the vector database and fine tuned and optimized in the later stage,could provide a generative dialogue and question answering service to assist the staff of the Financial Shared Service Center in conducting online and offline training work.Experimental verification shows that the system can accurately answer the questions raised by users,reduce the workload of training and promotion personnel,and also has application prospects in other railway professional training work.

关 键 词：文本大模型 培训 本地化部署 生成式对话 智能问答系统 

分 类 号：F530.68[经济管理—产业经济] TP39[自动化与计算机技术—计算机应用技术]