大数据 ›› 2019, Vol. 5 ›› Issue (6): 85-100.doi: 10.11959/j.issn.2096-0271.2019052

• 应用 • 上一篇    

W EB:一种基于网络嵌入的互联网借贷欺诈预测方法

王成1,2,3(),舒鹏飞1,2   

  1. 1 同济大学计算机科学与技术系,上海 201804
    2 嵌入式系统与服务计算教育部重点实验室,上海 201804
    3 上海智能科学与技术研究院,上海 200092
  • 出版日期:2019-11-15 发布日期:2020-01-10
  • 作者简介:王成(1980- ),男,同济大学计算机科学与技术系教授,主要研究方向为网络服务优化与安全、互联网金融反欺诈和网络空间异常事件检测研究|舒鹏飞(1994- ),男,同济大学计算机科学与技术系硕士生,主要研究方向为数据挖掘、机器学习和欺诈检测
  • 基金资助:
    国家自然科学基金资助项目(6197228);国家自然科学基金资助项目(61571331);工业和信息化部工业互联网创新发展工程项目([2018]282);霍英东教育基金会高等院校青年教师基金(151066);中央高校基本科研业务费专项资金资助项目(kx0137020181527);上海市青年拔尖人才开发计划

WEB:a fraud prediction method of Internet lending using network embedding

Cheng WANG1,2,3(),Pengfei SHU1,2   

  1. 1 Department of Computer Science,Tongji University,Shanghai 201804,China
    2 The Key Lab of Embedded System and Service Computing Ministry of Education,Shanghai 201804,China
    3 Shanghai Institute of Intelligent Science and Technology,Shanghai 200092,China
  • Online:2019-11-15 Published:2020-01-10
  • Supported by:
    The National Natural Science Foundation of China(6197228);The National Natural Science Foundation of China(61571331);The Major Project of Ministry of Industry and Information Technology of China([2018]282);Fok Ying-Tong Education Foundation for Young Teachers in the Higher Education Institutions of China(151066);The Fundamental Research Funds for Central Universities(kx0137020181527);Municipal Human Resources Development Program for Outstanding Young Talents in Shanghai

摘要:

基于关联图谱的互联网借贷欺诈预测方法限制了特征的挖掘效率、挖掘深度以及特征的可复用性、可表达性。针对此问题,引入网络嵌入技术,在保留欺诈特征的前提下,将网络中的节点嵌入低维的向量空间,利用向量对网络中的结构和语义信息进行表达;提出了基于周期性时间窗口的网络更新方法和决策批处理方法来提升网络嵌入在精准性和实时性方面的性能。实验表明,网络嵌入技术能够自动有效地学习网络中隐含的关联关系与特征;通过将传统方法和网络嵌入方法相结合,欺诈预测性能得到了显著提升。

关键词: 关联图谱, 互联网借贷, 网络嵌入, 反欺诈, 风险防控

Abstract:

Internet lending fraud prediction method based on association graph limits the mining efficiency and depth of features,as well as the reusability and expressibility of features.To solve this problem,the network embedding technology was introduced,and the structure and semantic information in the network by using the vector was expressed.The network update method based on periodic time window and decision batch method were proposed to improve the performance of network embedding in the two business requirements of accuracy and real-time.The experiment shows that the network embedding technology can automatically and effectively learn the implicit relationship and characteristics of the network.By combining the traditional method and the network embedding method,the fraud prediction performance has been significantly improved.

Key words: association graph, internet lending, network embedding, anti-fraud, risk management

中图分类号: 

No Suggested Reading articles found!