电信科学 ›› 2020, Vol. 36 ›› Issue (2): 52-60.doi: 10.11959/j.issn.1000-0801.2020048

• 研究与开发 • 上一篇    下一篇

基于协同过滤与社交网络混合算法的客户信用建模及授信方法

张科1,孙越佳1,韩海2   

  1. 1 中国移动通信有限公司研究院,北京 100053
    2 中国移动通信集团山西有限公司,山西 太原 030027
  • 修回日期:2020-02-11 出版日期:2020-02-20 发布日期:2020-05-19
  • 作者简介:张科(1981- ),男,中国移动通信有限公司研究院工程师,主要研究方向为电信客户信用建模、社交网络信息传播、机器学习、推荐系统等|孙越佳(1987- ),女,现就职于中国移动通信有限公司研究院,主要研究方向为电信客户信用建模、商务智能、机器学习等|韩海(1981- ),男,中国移动通信集团山西有限公司高级工程师、PMP,主要研究方向为互联网信息化、电子商务、信息系统项目管理

Customer credit modeling and credit granting method based on hybrid algorithm of collaborative filtering and social network

Ke ZHANG1,YueJia SUN1,Hai HAN2   

  1. 1 China Mobile Research Institute,Beijing 100053,China
    2 China Mobile Shanxi Co.,Ltd.,Taiyuan 030027,China
  • Revised:2020-02-11 Online:2020-02-20 Published:2020-05-19

摘要:

随着信用甄别市场需求的扩大,传统FICO信用建模方法依赖受信人大量历史信用行为、对群体属性识别能力弱的局限不断显现,造成了信用评分冷启动问题。提出了一种基于协同过滤与社交网络混合算法的信用评定模型,可有效解决信用评定冷启动困难的问题,在受信人信用评定数据不充足的情况下,亦可完成对其信用的相对准确评估。该算法首先基于受信人基本身份特征,通过协同过滤方法授予初始信用分,再通过社交网络信任图、群体聚类的学习成果修正信用评估模型。实验结果表明,该方法的信用预评估误差较低,可逐步引导预评分平稳过渡至正式评估。将该方法应用于信用培育期,可培养受信人良好的信用习惯,引导受信人信用行为良性循环。

关键词: 信用建模, 冷启动, 协同过滤, 社交网络, 融合算法

Abstract:

With the expansion of market demand for customer credit screening,the traditional FICO credit modeling method relies on a large number of historical credit behavior of trustees,and the limitation of weak ability to identify group attributes is constantly emerging,which causes the problem of cold start of credit scoring.Based on the identity characteristics of the trustee,the initial score through collaborative filtering method was firstly granted,and then the credit evaluation was modified through social network trust graph and group clustering learning results to solve the credit cold start problem during the transition period of scoring.The experimental results show that the credit pre-evaluation error of this method is low and can be smoothly transited to formal evaluation.This method can cultivate good credit habits of trustees in the initial stage of credit,guide the virtuous circle of credit behavior.

Key words: credit modeling, cold start, collaborative filtering, social network, fusion algorithm

中图分类号: 

No Suggested Reading articles found!