基于深度学习的银行间债券市场异常交易行为检测  

ABNORMAL BEHAVIOR DETECTION OF INTERBANK BOND MARKET BASED ON DEEP LEARNING

作　　者：黄良瑜 王薏婷 詹杭龙 金健[1] Huang Liangyu;Wang Yiting;Zhan Hanglong;Jin Jian(School of Computer Science and Technology,East China Normal University,Shanghai 200062,China;School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China;CFETS Information Technology(Shanghai)Co.,Ltd.,Shanghai 201203,China)

机构地区：[1]华东师范大学计算机科学与技术学院,上海200062 [2]上海大学计算机工程与科学学院,上海200444 [3]中汇信息技术(上海)有限公司,上海201203

出　　处：《计算机应用与软件》2021年第9期78-85,共8页Computer Applications and Software

基　　金：上海市科学技术委员会项目(18511103802,18511106202,18511103801)。

摘　　要：银行间债券市场作为金融市场重要组成部分,发挥着传导货币政策、提升资本流动性的作用。对市场异常交易行为的检测是保障银行间债券市场健康平稳运行、提升防范金融风险水平的有效手段。因此,提出一种基于网络嵌入和深度学习的异常交易行为检测方法,能有效检测出规则未知的异常交易行为。该方法结合交易网络的特点,采用一种面向时序属性网络的嵌入表示方法,并使用LSTM模型来检测异常交易行为。实验结果显示该模型F1指标值大于0.7,可提高异常交易行为检测模型的精确度。As an important part of the financial market,the inter-bank bond market plays a significant role in transmitting monetary policy and improving capital mobility.The detection of abnormal market trading behavior is an effective means to ensure the healthy and stable operation of the inter-bank bond market and improve the level of prevention of financial risks.Therefore,this paper proposes an abnormal trading behavior detection method based on network embedding and deep learning,which can detect abnormal transaction behaviors with unknown rules effectively.Combined with the characteristics of the inter-bank bond market trading network,an embedding representation method for time-series attribute networks was adopted.The LSTM model was then used to detect abnormal trading behavior.The experimental results show that the F1 score of this model is over 0.7,and this method can improve the accuracy of the transaction behavior detection model.

关 键 词：银行间债券市场 网络嵌入 长短期记忆网络 异常检测 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]