文献类型：期刊文献

中文题名：LightRAG框架下AI答疑系统的设计与实现

英文题名：Development of Intelligent Q&A System Based on LightRAG Framework

作者：罗小娟[1];程奕豪[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2026

卷号：45

期号：1

起止页码：8

中文期刊名：实验室研究与探索

外文期刊名：Research and Exploration In Laboratory

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金资助项目(61872143);华东理工大学2025年本科教育教学改革项目(校教[2024]29号)。

语种：中文

中文关键词：AI答疑系统;LightRAG框架;检索增强生成;教育人工智能

外文关键词：AI Q&A system;LightRAG framework;retrieval enhancement generation;educational artificial intelligence

摘要：针对传统教育答疑效率低下以及现有大模型易产生幻觉的问题,构建了一种基于检索增强生成(RAG)技术的智能答疑系统,旨在提升知识检索准确性与答疑响应效率。该系统采用LightRAG框架,融合知识图谱与向量数据库构建高效检索模块,并通过增量更新机制保障知识的时效性。以“电路原理”课程为应用案例,优化了知识库构建流程,完成了嵌入模型筛选与提示词模板改进;引入基于RoBERTa架构的重排序模型,对检索结果进行语义层面的精准排序;基于Gradio搭建可视化交互平台,支持多模式查询与知识库信息的可视化展示。实验验证表明,该系统能够有效降低大模型的幻觉发生率,且具备响应迅速、答案准确、快速适配知识更新迭代的优势:与现有基于GraphRAG的系统相比,其答疑响应时间平均缩短6%,显著提升了学习效率。此外,系统创新性融合了多源检索与语义重排序技术,大幅提高了教育问答的可靠性、具备良好的推广应用价值。
Aiming at the problems of low efficiency of traditional education Q&A and hallucinations of large models,this study aims to construct an intelligent Q&A system based on RAG technology to improve the accuracy and response efficiency of knowledge retrieval.By using the LightRAG framework,the knowledge graph and vector database are integrated to achieve efficient retrieval,and the incremental update mechanism is used to ensure the timeliness of knowledge.Taking the circuit principle course as an example,the knowledge base construction process is optimized,the embedded model is screened and the prompt word template is improved.The reranking model based on RoBERTa architecture is introduced to semantically rank the retrieval results.Finally,a visualization platform based on Gradio is built to support multi-mode query and knowledge base information visualization.The system can effectively alleviate the illusion of model,respond quickly and answer accurately,and can quickly adapt to the update iteration of knowledge,which greatly improves the learning efficiency.Compared to the GraphRAG system,the response time for answering questions reduces by an average of approximately 6%.The system innovatively integrates multi-source retrieval and semantic re-ranking technology,improves the reliability and efficiency of educational question answering,and has good application value.