基于混合检索与提示词优化的增强生成框架及其应用

打开文本图片集

中图分类号：TP391.1 文献标识码：A 文章编号：1006-8228（2026）01-40-07

Abstract：Retrieval-augmentedgeneration（RAG）methodshavebeenshowntosignificantlyenhancelargelanguagemodels'ability toutlizextealkowledgeandimproegenerationqualitypartiularlyindomain-specificquestio-answeringsenariosispaper addresesdeficienciesinretrievalandgenerationperformanceoflargelanguagemodelsformathematicsteachingquestionanswering bydesigningandimplementinganimprovedRAGframeworkapliedtoamathematicsQAteachingsystem.Theproposed frameworkintegratesahybridretrievalstrategythatcombinesBM25withdense-vectorretrievaltoimprovebothrecalland precisiooftheretriever，andintroducesaPromptEngineringstrategytoenhancethequalityofthemode'sgeneratedaswers. Experimentalresultsdemonstratethattheproposedframeworkreliablyperformstheend-toendrtrieval-togenerationprocessin mathematics QA tasks;the retriever achieves a Hit Rate @10 of 77.77% ，validatingthe feasibility and effectivenessof the method indomain-specificQAaplicationsandprovidingapracticalreferencefordesigningteachingQAsystemsinotherdisciplines.

Keywords：Retrieval-AugmentedGeneration；HybridRetrieval;PromptEngnering;LargeLanguageModel;MathematicalQuestion Answering

0引言

近年来，大语言模型（LargeLanguageModel，LLM）的快速发展推动了智能问答系统在教育、搜索与知识服务等领域的广泛应用。（剩余11034字）