基于法条检索的生成式法律问答研究

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：0253-2395（2025）04-0653-13

Abstract：Pre-tranedlanguagemodel-basedlegalquestionansweringsystemsstruggletoflexiblyunderstandusersintentandlack the integrationofexteralknowledge，making itdificulttoachievedesiredresults.Toadressthis，thispaper proposesafinegrainedegalquestionansweringdatasetbasedonthecriminallawarticleslibrary（FCL-QA）.BasedonFCL-QA，thispaperproposes aStatutoryArticles Retrieval AugmentedQuestion AnsweringFramework （SaRAF）basedonlarge language model.Thecoreidea is tolocateteategoryofteuestintroughulti-vellasiicationothoeofatutoryarticlesthoughtategory tofacilitateretrieval，andfiallygenerate theanswerusingalargelanguage model.ExperimentalresultssowthatteSaFoutperformsbothwithouttatutoryarticlesgenerationmethodandRetrieval-augmentedGeneration（RAG）method，achievingROUGE⊥F1 score of 42.27% ，BLEU-4 score of 27.78% and BERTScore of 72.52% on the FCL-QA dataset. Key words： large language model; question answering system; retrieval augmented

0 引言

法律问答的目标是为用户提供高质量、高可靠的法律咨询，是自然语言处理技术在司法领域的重要应用。（剩余17827字）