基于IARPA项目指南的智能情报技术解析与布局启示  

Parsing and Layout Insights of Agentic Document and Information Service Technology Based on IARPAs Project Announcements

作　　者：付芸 刘细文[1,2] Fu Yun;Liu Xiwen(National Science Library,Chinese Academy of Sciences,Beijing 100190;Department of Information Resources Management,School of Economics and Management,University of Chinese Academy of Sciences,Beijing 100190)

机构地区：[1]中国科学院文献情报中心,北京100190 [2]中国科学院大学经济与管理学院信息资源管理系,北京100190

出　　处：《情报学报》2025年第3期257-270,共14页Journal of the China Society for Scientific and Technical Information

基　　金：国家社会科学基金重大项目“数智转型背景下智能情报关键技术应用研究”(23&ZD228)。

摘　　要：明确智能情报技术的内涵和外延对情报机构规划和科研人员选题至关重要。本文探索使用美国情报高级研究计划局发布的87个项目指南全文本开展细粒度技术内容分析,从主题-子主题-技术问题-评估指标四个层级维度立体化揭示其围绕智能情报技术的布局内容和特征,在此基础上总结得出智能情报技术是什么、包含哪些内容。为此,本文提出基于项目指南的技术布局分析框架,设计项目指南描述模型(project description model,PDM),包含研究目标、技术问题和评估指标三类知识元素;研发适用于PDM的项目指南文本解析提示词,在GPT-4o(generative pretrained transformer 4 omni)中应用后三类知识元素平均识别准确率达92.94%;使用TopicGPT依次处理项目指南的概要文本和研究目标文本并构建分层主题(主题-子主题);使用分层主题、技术问题与评估指标揭示技术布局内容和特征。经分析得出智能情报技术是指应用于情报工作的智能技术与升级情报工作流程智能化的信息技术,涵盖智能数据、智能计算、智能认知和智能系统四个方面,具有四大关键特征和四类评估原则,相关案例佐证了本文结论的可靠性。最后,在大情报观指导下,本文提出面向未来智能情报技术布局的六大任务方向,包括三类关键技术研发问题:情报应用情景的智能解析与认知建模技术、目标驱动的跨时空模态情报数据生产与组织技术、适用于复杂情报情景计算与分析的智能信息技术;三类核心实践应用问题:情报工作流程的标准化与自动化、情报工具的易用性与分析结果的可传播性、情报工作评估体系的确定性与可测量性。Clarifying the connotation and scope of Agentic Document and Information Service Technology(ADIST)is crucial for the strategic planning of Document and Information Service(DIS)institutes and the research direction of scholars.This study explores a fine-grained technical content analysis of 87 project announcements published by IARPA to systematically reveal the structure and characteristics of ADIST across four hierarchical dimensions:topic,subtopic,technical issue,and evaluation metric.Accordingly,this study identifies what ADIST is and its key components.To achieve this,a technology layout analysis framework based on project announcements was proposed.A Project Description Model(PDM)was designed incorporating three categories of knowledge elements:research objectives,technical questions,and evaluation metrics.A project announcement text-parsing prompt tailored for PDM was developed,achieving an average recognition accuracy of 92.94%for the three knowledge elements when applied in GPT-4o(Generative Pretrained Transformer 4 Omni).Additionally,TopicGPT was used to construct hierarchical topics(topic-subtopic)based on the summary and research objective texts of project announcements.This study further reveals the technical layout content and characteristics by integrating hierarchical topics,technical issues,and evaluation metrics.The analysis concludes that ADIST refers to intelligent technologies applied to DIS tasks and information technologies that enhance the agentic-driven transformation of the DIS workflow.It encompasses four key aspects—intelligent data,intelligent computing,intelligent cognition,and intelligent systems—with four core characteristics and four evaluation principles.Case studies further validate the reliability of these findings.Finally,guided by a broad DIS perspective,this study proposes six key future-oriented tasks for ADIST development.These include three key technical research challenges:intelligent analysis and cognitive modeling for DIS application scenarios,goal-driven multimodal

关 键 词：智能情报技术 技术布局 项目指南 项目指南描述模型 

分 类 号：G250.2[文化科学—图书馆学]