融入标签信息的民间借贷案情要素识别方法  

Private loan case factor recognition method incorporated with label information

作　　者：左江涛 张恒滔 ZUO Jiangtao;ZHANG Hengtao(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Yunnan Key Laboratory of Artificial Intelligence,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500 [2]云南省人工智能重点实验室,云南昆明650500

出　　处：《现代电子技术》2023年第21期76-81,共6页Modern Electronics Technique

基　　金：云南省基础研究计划(202001AS070014)。

摘　　要：民间借贷案情要素识别旨在通过对裁判文书的理解,将能够反映案件主要事实的要素按照属性体系提取出来。现有的要素识别主要基于序列标注方法进行,但民间借贷案情要素包含的部分隐式要素无法被标注。文中将其转化为多标签文本分类任务,通过“民事BERT”对标签属性、要素标签与裁判文书事实描述进行语义编码,基于前馈神经网络构建的融合模块将事实描述表示分别和预定义的十类要素属性表示进行特征融合,再利用标签注意力机制获得某一要素属性下不同要素标签的关注度,最后借助分类器分别识别每一类属性的要素。实验结果显示,该方法在验证集和测试集上的平均F1值较基线模型BERT均有提升。The recognition of private loan case factors aims to extract the factors that can reflect the main facts of the case according to the attribute system after thorough understanding the judgment documents.The existing factor recognition is mainly based on the sequence labeling method,but some implicit factors contained in the private loan case will be overlooked.In this paper,those will be transformed into a multi⁃label text classification task.The″civil BERT″is used to semantically encode the label attributes,factor labels and fact description of the referee documents.The fusion module based on the feed⁃forward neural network implements feature fusion on the fact description representations and the predefined ten factor attribute representations.And then,the label attention mechanism is used to obtain the attention degree of different factor labeling for a certain factor attribute.Finally,the factors of each type of attribute are identified by classifiers.The experimental results show that the average F1 value of the method on the verification set and test set is higher than that of the baseline model BERT.

关 键 词：智慧司法 案件要素识别 多标签文本分类 民间借贷案件 自然语言处理 注意力 

分 类 号：TN911.7-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]