互联网金融安全情绪感知及风险预警应用研究——基于BERT所作的探索  

Research on Internet Financial Security Sentiment Perception with Applications in Risk Early Warning

作　　者：丁晓蔚[1,2,3] 季婧 赵笑宇 王本强 丁毅杰 王献东 Ding Xiaowei;Ji Jing;Zhao Xiaoyu;Wang Benqiang;Ding Yijie;Wang Xiandong(School of Information Management,Nanjing University,Nanjing 210023;Nanjing University Research Center for Inclusive&Rural Financial Technology Innovation,Nanjing 210023;Zijin Media Think Tank,Nanjing 210023;Shanghai Midu Information Technology Co.,Ltd.,Shanghai 201024)

机构地区：[1]南京大学信息管理学院,南京210023 [2]南京大学普惠三农金融科技创新研究中心,南京210023 [3]紫金传媒智库,南京210023 [4]上海蜜度信息技术有限公司,上海201204

出　　处：《情报杂志》2023年第9期57-70,共14页Journal of Intelligence

基　　金：国家社会科学基金一般项目“基于大数据的情报分析在互联网金融风险防控场景中的应用研究”(编号:19BTQ070)的阶段性成果。

摘　　要：[研究目的]在为互联网金融(以下简称“互金”)风控监管构建风险情报支撑体系的视域中,进行专门面向互金安全情绪感知技术的研究,提出并构建基于BERT的互金安全情绪感知模型。[研究方法]基于BERT分别构建互金安全情绪感知模型(IFSSP)、集成指数移动平均EMA的互金安全情绪感知模型(IFSSP+EMA)、集成指数移动平均EMA+对抗训练PGD的互金安全情绪感知模型(IFSSP+EMA+PGD),并将其用于从高维、异构、稀疏、高噪、高频的海量大数据中,快速、准确地挖掘隐藏信息和安全线索,提取出风险预警因子、特征和指标,并进一步测试其对风险情报体系的效果。最后进行微观、中观、宏观应用实证研究。[研究结论]基于BERT的面向互金安全情绪感知技术能提高互金风险预警的准确率,在应用实证中体现出可行性和有效性。[Research purpose]In the perspective of building a risk intelligence support system for Internet finance risk control and supervision,we conduct research on finance security sentiment perception technology specially oriented for Internet finance security.We propose and build an Internet finance security sentiment perception model based on BERT.[Research method]Based on BERT,the Internet finance security sentiment perception model(IFSSP),the Internet finance security sentiment perception model integrating exponential moving average EMA(IFSSP+EMA),and Internet finance security sentiment perception model integrating exponential moving average EMA and Adversarial Training PGD(IFSSP+EMA+PGD)were constructed respectively,and used to timely and accurately mine hidden information and security clues from high-dimensional,heterogeneous,sparse,high noise,and high-frequency massive big data,extract early-warning risk factors,characteristics and indicators,and further test their effects on the risk intelligence system.Finally,empirical researches on micro,meso and macro applications were carried out.[Research conclusion]Internet Finance Security Sentiment Perception technology based on BERT can improve the accuracy of Internet Financial risk early-warning,which shows its feasibility and effectiveness in empirical applications.

关 键 词：互联网金融 数字金融 风险情报体系 互金安全情绪感知 BERT 金融风险预警技术 国家金融安全 

分 类 号：G350[文化科学—情报学] F830-4[经济管理—金融学]