文献类型：期刊文献

中文题名：基于知识图谱和RAG技术的古诗词智能问答系统

英文题名：An Intelligent Question-Answering System for Classical Chinese Poems Based on Knowledge Graph and RAG Technology

作者：翁嘉屹[1];翟洁[1];熊瀚锐[1];蒋奕辰[1];张玥[1];耿俊伟[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2025

卷号：9

期号：20

起止页码：115

中文期刊名：现代信息科技

外文期刊名：Modern Information Technology

语种：中文

中文关键词：幻觉问题;检索增强生成;自然语言处理;知识图谱;大语言模型;命名实体识别

外文关键词：hallucination problem;Retrieval-Augmented Generation;Natural Language Processing;Knowledge Graph;Large Language Model;Named Entity Recognition

摘要：针对大模型在古诗词领域普遍存在的作者混淆、朝代错位等幻觉问题,本项目创新性地提出了基于RAG(检索增强生成)技术的解决方案。通过构建专业古诗词知识图谱,并设计多维度提示工程,系统能够有效纠正大模型在诗词背景、历史脉络等复杂语义理解上的错误。具体实现上,首先利用命名实体识别提取问题关键信息,再通过知识图谱检索获取权威内容作为生成依据,最后经RAGas评估验证,大模型幻觉率得到显著降低。这一研究填补了大模型在传统文化领域应用的准确性空白,为AI赋能文化传承提供了可靠的技术路径。
In view of the hallucination problems such as author confusion and dynasty dislocation in the field of ancient poetry,this project innovatively proposes a solution based on Retrieval-Augmented Generation(RAG)technology.By constructing a professional Knowledge Graph of ancient poetry and designing multi-dimensional prompt engineering,the system can effectively correct the errors of the Large Language Model in the complex semantic understanding of poetry background,historical context and other aspects.In terms of specific implementation,firstly,Named Entity Recognition is used to extract the key information of the problem,and then authoritative content is obtained through Knowledge Graph retrieval as the basis for generation.Finally,RAGas evaluation verifies that the hallucination rate of the Large Language Model is significantly reduced.This study fills the gap in the accuracy of the application of the Large Language Model in the field of traditional culture and provides a reliable technical path for AI to enable cultural inheritance.