融合大语言模型的情报智库政策内容问答服务研究——以粮食安全政策为例  

Research on Policy Content Q&A Service of Intelligence Think Tank Integrating Large Language Models--A Case Study of Food Security Policy

作　　者：刘彦辉[1] 张海涛[1,2,3] 周红磊 庞宇飞 Liu Yanhui;Zhang Haitao;Zhou Honglei

机构地区：[1]吉林大学商学与管理学院,吉林长春130012 [2]吉林大学信息资源研究中心,吉林长春130012 [3]吉林大学国家发展与安全研究院,吉林长春130012

出　　处：《图书与情报》2025年第1期9-20,共12页Library & Information

基　　金：国家社会科学基金重大项目“总体国家安全观下重大突发事件的智能决策情报体系研究”(项目编号:20&ZD125)研究成果之一。

摘　　要：2025年初,我国发布的DeepSeek-R1推理模型,促进了普惠化AI时代的到来,如何将智能技术有效融入情报智库以提升其服务响应能力,并基于特定领域数据进行服务场景的验证,实现理论到实践的有效衔接,已成为情报智库实践中亟待解决的关键问题。文章聚焦于以大语言模型赋能情报智库政策内容问答服务,首先分析了情报智库的政策内容问答服务任务,探讨了大语言模型在该场景下的可行性;其次设计了融合大语言模型的情报智库粮食安全政策内容问答服务方案;最后以粮食安全政策文本为核心数据源,对方案进行了实证研究。通过从政策文本提取要素与要素关系构建数据集,采用LoRA方法微调DeepSeek模型,将其接入LangChain框架,同时挂载本地知识库,形成完整的服务方案。结果表明,该方案具有较高的可行性和专业性,为情报智库在特定领域实现政策内容问答的场景服务提供了重要的理论依据与实践支持。In early 2025,China released the DeepSeek-R1 inference model,promoting the arrival of an era of AI that benefits all.How to effectively integrate intelligent technologies into intelligence think tanks to enhance their service response capabilities,and validate service scenarios based on data from specific fields to achieve a seamless connection between theory and practice,has become a key issue that needs to be addressed in the practical application of intelligence think tanks.This study focuses on empowering policy content Q&A services in intelligence think tanks using large language models.First,it analyzes the tasks involved in policy content Q&A services and explores the feasibility of large language models in this context.Next,a service solution that integrates large language models for food security policy Q&A in intelligence think tanks is designed.Finally,empirical research is conducted using food security policy texts as the primary data source.The study constructs datasets by extracting elements and their relationships from policy texts,fine-tunes Deepdeek model using the LoRA method,integrates it into the LangChain framework,and incorporates a local knowledge base to form a complete service solution.The results demonstrate the high feasibility and professionalism of the proposed solution,providing essential theoretical and practical support for the implementation of policy content Q&A services in specific domains within intelligence think tanks.

关 键 词：大语言模型 粮食安全 政策内容问答 情报智库服务 知识库 

分 类 号：G351.3[文化科学—情报学]