大语言模型在法律事件检测中的效能评估

打开文本图片集

DOI：10.16644/j.cnki.cn33-1094/tp.2026.01.011

中图分类号：D90-055；TP18 文献标识码：A 文章编号：1006-8228（2026）01-58-05

Abstract：Generativeartifiialintellgene，particularlylargelanguagemodels（Ls），isinreasinglyaldinthlegaldomain. However，accuratelyextractingeventsfromcourtjudgmentstoconstructstructuredfactualinformationremainschalengingThis paperexplorestheeficacyofLLMsinChineselegaleventdetectiontass，hypothesizingtheirabilitytoefectivelyidentifynd anotatekeyeventswithincourtjudgments.Tovalidatethishypothesis，thestudyconductedzero-shoteventextractionexperiments usingtheLEVENdataset withfourmodels：DeepSek-V3，DepSeek-R1，Doubao-See-1.6，andQwen-Plus.Resultsindicatehigh overallprecisionoftheselectedmodels，thoughrecallandstabiltyrequireimprovement.Costanalysisrevealstheselected models'significantlylowertotalexecutionexpensescomparedtomanualanotation，demonstratingsubstantialeconomicadvantages. Thisstudyconfirmsthepotentialoflargelanguagemodelsinlegaleventdetection，providinganempiricalfoundationfor constructing high-quality event databases and enhancing judicial reasoning transparency.

Keywords：Legal Event;LargeLanguage Models；Event Detection;Artificial Intelligence

0引言

大语言模型的兴起，推动人工智能在自然语言理解与生成领域实现能力飞跃。（剩余7529字）