考虑审计要素多重语义关联的财务欺诈识别  

Financial statement fraud identification considering the multiple-dimensional semantic associations of auditing elements

作　　者：李建平 孙灏 常闫芃 朱晓谦 LI Jian-ping;SUN Hao;CHANG Yan-peng;ZHU Xiao-qian(School of Economics and Management,University of Chinese Academy of Sciences,Beijing 100190,China;Institutes of Science and Development,Chinese Academy of Sciences,Beijing 100190,China;School of Public Policy and Management,University of Chinese Academy of Sciences,Beijing 100049,China;MOE Social Science Laboratory of Digital Economic Forecasts and Policy Simulation at UCAS,Beijing 100190,China)

机构地区：[1]中国科学院大学经济与管理学院,北京100190 [2]中国科学院科技战略咨询研究院,北京100190 [3]中国科学院大学公共政策与管理学院,北京100049 [4]中国科学院大学数字经济监测预测预警与政策仿真教育部哲学社会科学实验室(培育),北京100190

出　　处：《管理科学学报》2024年第3期58-70,共13页Journal of Management Sciences in China

基　　金：国家自然科学基金资助重点项目(92046023);国家自然科学基金资助项目(72371236,71971207);中央高校基本科研业务费专项资金;中国科学院大学数字经济监测预测预警与政策仿真教育部哲学社会科学实验室(培育)基金.

摘　　要：现有的财务欺诈识别研究大多基于公司、审计师、会计师事务所等审计要素中较为简单的关系特征,罕有研究能够系统刻画各类审计要素之间错综复杂的关联关系.本文创新性地引入知识图谱(Knowledge Graph)技术,构建出包含公司、审计师和会计师事务所的多重语义关联网络,并利用图神经网络(Graph Neural Networks)方法捕捉知识图谱中审计要素之间复杂隐秘的关联关系以提高财务欺诈识别效果.基于我国2018年—2019年的上市公司样本,构建出包含12373个审计要素和111194条关系的审计知识图谱.实证研究发现引入审计要素关联关系能够提升财务欺诈识别准确率;在多种审计要素关联关系中,考虑审计师对公司出具的审计意见对欺诈识别更为重要;对比不同历史时长的审计要素,使用公司历史5年的审计要素识别财务欺诈的效果更好.本研究可以为投资者、分析师以及监管机构在大数据时代下的财务欺诈识别提供科学参考.Existing studies on financial fraud identification mainly focus on analyzing relatively simple features of audit elements such as companies,auditors,and audit firms.Few studies can systematically describe the intricate relationships between various auditing elements.To improve the accuracy of financial fraud identification,this paper innovatively introduces the knowledge graph technology to construct a multi-dimensional semantic network including companies,auditors,and audit firms,and applies graph neural network models to analyze the complex relationships among the auditing elements in the knowledge graph to better detect financial fraud.Based on the auditing information of listed companies in China from 2018 to 2019,an auditing knowledge graph containing 12373 auditing elements and their 111194 relationships is constructed.The empirical results demonstrate that the relationships among auditing elements can indeed help improve the accuracy of financial fraud identification.Among the various relationships,the auditing opinion of auditors to companies is more important for fraud identification.Using the auditing element information of a company in the past 5 years to identify fraud is better than in other historical periods.This study can provide a scientific reference for investors,analysts,and regulators to identify financial fraud in the era of big data.

关 键 词：财务欺诈 审计要素关系 语义关联 知识图谱 图神经网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] F830[自动化与计算机技术—控制科学与工程]