基于深度学习的企业信用风险画像构建方法研究  

Research on the construction method of corporate credit risk portrait based on deep learning

作　　者：付学敬 马宏文 袁晓月 Fu Xuejing;Ma Hongwen;Yuan Xiaoyue(Information Application Research Center of Shanghai Administration for Market Regulation,Shanghai 200032,China)

机构地区：[1]上海市市场监督管理局信息应用研究中心,上海200032

出　　处：《现代计算机》2024年第16期30-36,共7页Modern Computer

基　　金：国家市场监督管理总局技术保障专项(2022YJ29);国家市场监督管理总局科技计划项目(2023MK035);上海市市场监督管理局年度科技项目(2023-56)。

摘　　要：为更好地刻画企业信用风险特征,研究提出基于数据分布与深度学习的企业信用风险画像构建方法。首先,根据企业信用风险模型构建企业信用风险标签体系,着力刻画空间维度的群体比较特征和时间维度的个体变化特征。然后,针对统计类标签,按照特定规则定义产生特征标签;针对挖掘/统计类标签,根据具体任务采用命名实体识别和情感分类等深度学习技术构建特征标签。企业信用风险画像能够直观揭示企业信用风险特征,深度学习技术有力拓展了企业特征刻画能力,更好服务于信用监管和智慧监管。To better characterize corporate credit risk characteristics,this study proposes a method for constructing corporate credit risk portraits based on data distribution and deep learning.First,an enterprise credit risk label system is constructed based on the enterprise credit risk model,focusing on characterizing the group comparison characteristics in the spatial dimension and the individual change characteristics in the time dimension.Then,for statistical labels,feature labels are generated according to specific rule definitions;for mining/statistical labels,deep learning technologies such as named entity recognition and emotion classification are used to construct feature labels based on specific tasks.Corporate credit risk portraits can intuitively reveal corporate credit risk characteristics.Deep learning technology has effectively expanded corporate characteristics characterization capabilities and better served credit regulation and smart regulation.

关 键 词：深度学习 企业画像 数据分布 信用风险 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] F270[自动化与计算机技术—控制科学与工程]