O'Reilly、Cloudera 主办
Make Data Work
2017年7月12-13日:培训
2017年7月13-15日:会议
北京,中国

大数据时代银行客户社交关系圈研究与应用 (Research on and the application of a social relation circle of bank customers in the big data era)

此演讲使用中文 (This will be presented in Chinese)

黄文宇 (广发银行股份有限公司)
14:50–15:30 Friday, 2017-07-14
企业应用 (Enterprise adoption)
地点: 多功能厅5B+C(Function Room 5B+C) 观众水平 (Level): 中级 (Intermediate)
平均得分:: ****.
(4.50, 2 次得分)

必要预备知识 (Prerequisite Knowledge)

银行基础知识、算法建模知识

您将学到什么 (What you'll learn)

社群发现算法应用场景、传统银行业务场景

描述 (Description)

为加深对银行客户的洞察,提升银行营销获客与风险管控能力,广发银行基于Hadoop大数据平台,通过Hive on Spark、图计算进行数据加工,结合LFM社群发现、增强决策树等机器学习算法构建了银行客户社交关系模型,挖掘出银行客户社交关系圈,并应用于银行实际业务中。银行客户社交关系圈全面的反映了银行个人客户资金、社交等关系,以全新的视角实现银行对客户洞察从点到面、从单客到客群的扩展,填补银行个人客户社交关系研究与应用的空白。
银行客户社交关系圈简要处理流程如下:首先,银行整合加工了近6000万存量个人客户及1亿个潜在个人客户的一年内交易流水及其他关系数据,提取资金、社交、媒介及位置四类关系;其次,选择以LFM社群划分算法为基础,设计了权重、导率、归属度等指标,识别、挖掘客户的个人圈及客群圈;最后,通过回归算法进行客户模式识别,形成客户分类,并在银行零售业务场景进行应用与推广。
目前广发银行以银行客户资金关系圈为基础,在信用卡及网络金融业务的失联催收、精准营销等方面开展与推广:
失联催收客户联系率51.57%,其中银行客户资金关系圈作为唯一可联信息来源的客户最高占比10.07%,预计每年可增加失联催收回款900万。
信用卡分期业务营销成功率提升15.67%,业务收益增加189万元,预计每年增加分期收益约4500万元。


CGB conducts data processing through Hive on Spark and graph computing in order to deepen its understanding of customers and enhance its ability of marketing and risk control. Combined with LFM community discovery, enhanced decision trees, and other machine learning algorithms, CGB creates social relation model for its customers, discovers their social relationship circle, and then applies this discovery to the bank’s real businesses. A customer’s social relations circle fully reflects their personal funds and social and other relationships and builds customer insights from a single customer to customer groups, filling in the gap between research on customers’ social relations and its application in the real world.

黄文宇 outlines CGB’s procedure for processing its bank customers’ social relation circles: First, CGB integrated and processed a year of transaction logs and other relation data about nearly 60 million existing customers and 100 million potential customers, extracting four types of relationships: capital, social, media, and location. Secondly, the bank used the LFM-based community segmentation algorithm, designed the weight, conductivity, attributes, and other matrices, and identified and mined a customer’s individual relation circle and group circle. Finally, CGB used regression algorithms to discover customer patterns, classify customer groups, and apply and promote their uses in the bank’s retail business scenarios.

Based on customers’ capital relation circles, currently GCB is launching and promoting services from dunning lost customers in credit card and online finance services to precision marketing. The touch rate is 51.57% for dunning to lost customers (of which the customer financial relation circle as the only source of connect information can be 10.07% at the highest proportion) and is expected to bring an increase of 9 million RMB dunning payback each year. The marketing success rate of credit card installment service increased by 15.67% (business income increased by 1.89 million RMB) and is expected to increase the annual income of installment payments by about 45 million RMB.

Photo of 黄文宇

黄文宇

广发银行股份有限公司

工学博士,现任广发银行数据中心总经理,银行业信息科技发展与风险管理专家,广州市金融高级专业人才,曾任工商银行北京数据中心信息科技专家。黄先生作为银行业科技条线的资深专家,在基础设施与运行维护、信息科技治理与管理、大数据研究及规划等领域具有丰富经验。

联系OReillyData

关注OReillyData微信号获取最新会议信息并浏览前沿数据文章。

WeChat QRcode

 

Stay Connected Image 1
Stay Connected Image 3
Stay Connected Image 2

阅读关于大数据的最新理念。

ORB Data Site