中国标准化Issue(2)：43-47,5.DOI:10.3969/j.issn.1002-5944.2026.02.002
基于检索增强生成与对话历史管理的标准智能问答系统研究
Research on an Intelligent Q&A System for Standards Based on Retrieval-augmented Generation and Dialogue History Management
摘要
Abstract
To address the challenges in professional standard Q&A,such as high demands for answer accuracy,strong contextual dependencies,and personalized user needs,this paper develops an intelligent Q&A system based on a retrieval-augmented generation(RAG)architecture integrated with dialogue history management.The system employs a constructed standard semantic knowledge base as a precise knowledge source and adopts a dual-path"retrieval-generation"strategy:First,dense vector retrieval technology is used to quickly recall the most relevant standard clauses or knowledge subgraphs from the knowledge base.Then,the retrieved precise knowledge fragments,along with the original user query,are fed into a generative model pre-trained on large-scale standard texts and fine-tuned,to produce clear,fluent natural language answers,effectively ensuring both accuracy and readability.Furthermore,the system innovatively incorporates a dialogue history management module based on topic modeling and sequence encoding,which dynamically analyzes the objects,topics,and structure of user conversations,enabling intelligent storage and contextual retrieval of historical dialogues.This equips the system with multi-turn,coherent conversational capabilities.Application validation in the context of oilfield safety and environmental standards demonstrates that the system achieves high accuracy in single-turn Q&A and effectively resolves references and contextual understanding in multi-turn interactions,significantly enhancing the intelligence level of standard knowledge services.关键词
智能问答/检索增强生成/预训练语言模型/对话管理/标准数字化/人机交互Key words
intelligent Q&A/retrieval-augmented generation(RAG)/pre-trained language model/dialogue management/standards digitalization/human-computer interaction引用本文复制引用
甘克勤,高亮,肖宝坤,林良红..基于检索增强生成与对话历史管理的标准智能问答系统研究[J].中国标准化,2026,(2):43-47,5.基金项目
本文受中国标准化研究院基本科研业务费项目"基于标准语义知识的智能问答关键技术应用研究"(项目编号:252024Y-11459)资助. （项目编号:252024Y-11459）
