摘要
【目的】鉴于司法裁判文书摘要需要与原文在基于案件事实、法律适用等要素方面保持一致,提出嵌入司法要素事实一致性评测的中文司法裁判文书摘要生成方法。【方法】定义司法裁判文书摘要事实一致性判定的原则和方法;确定数据增加、事实一致性纠错和测评等预处理流程;分别构建分段抽取模型和引入司法要素知识图的生成式摘要模型,并在CAIL2020数据集上进行实验。【结果】本文提出的FC-JDSM模型生成的摘要在指标ROUGE-N(N=1、2、L)、SRO、EM-FCJS上分别为67.98%、55.40%、64.14%、78.54%、90.01%,均优于比较模型。消融实验证实了分块抽取和事实信息引入的有效性。【局限】事实一致性评测模型中的数据增强方案得到的数据与真实数据存在偏差。【结论】将司法要素融入一致性评测和摘要生成过程中,能提高中文司法裁判文书摘要一致性,有利于司法工作的公正性。
[Objective]Summaries of legal judgment documents must remain consistent with the original text in terms of case facts and legal applications.This study proposes a method for summarizing Chinese legal judgment documents that incorporates factual consistency evaluation based on judicial elements.[Methods]First,we defined the principles and methods for determining factual consistency in legal summaries and established preprocessing steps,including data augmentation,consistency correction,and evaluation.Then,we built a segmented extraction model and a generative summarization model incorporating a judicial element knowledge graph.Finally,we conducted experiments on the CAIL2020 dataset.[Results]The proposed FC-JDSM model achieves performance metrics of ROUGE-N(N=1,2,L):67.98%,55.40%,64.14%;SRO:78.5%;and EM-FCJS:90.01%,all outperforming baseline models.The ablation experiments confirm the effectiveness of block extraction and factual information introduction.[Limitations]The data generated through the data augmentation method for the factual consistency evaluation model may deviate from real-world data.[Conclusions]Incorporating judicial elements into the consistency assessment and abstract generation process enhances the consistency of Chinese legal judgment instrument summaries and promotes fairness of judicial work.
作者
向博文
柴梦丹
向卓元
Xiang Bowen;Chai Mengdan;Xiang Zhuoyuan(Procurement and Bidding Management Center,Zhongnan University of Economics and Law,Wuhan 430073,China;School of Information Engineering,Zhongnan University of Economics and Law,Wuhan 430073,China)
出处
《数据分析与知识发现》
北大核心
2025年第8期73-85,共13页
Data Analysis and Knowledge Discovery
基金
国家自然科学基金项目(项目编号:72374219)的研究成果之一。
关键词
司法文书
自动摘要
事实一致性
评测指标
摘要生成模型
Judicial Documents
Automatic Summarization
Factual Consistency
Assessment Indicators
Summary Generation Model
