周静,周小宇,王汉生.自我网络特征对电信客户流失的影响[J].管理科学,2017,30(5):28-37
自我网络特征对电信客户流失的影响
On the Influence of Ego Network Concerning Customer Attrition of the Telecommunication Industry
投稿时间:2016-11-13  修订日期:2017-07-20
DOI:
中文关键词:  社交网络    信息熵  客户流失  自我网络
英文关键词:social network  degree  entropy  customer attrition  ego network
基金项目:中国人民大学科学研究基金重大规划项目《互联网统计学研究》资助
作者单位E-mail
周静 中国人民大学 统计学院北京 100872 zhoujing_89@126.com 
周小宇 上海科技大学 创业与管理学院上海 201210 zhouxy@shanghaitech.edu.cn 
王汉生 北京大学 光华管理学院北京 100871 hansheng@pku.edu.cn 
摘要点击次数: 18
全文下载次数: 
中文摘要:
      近年来,随着移动通信行业的蓬勃发展,市场饱和度越来越高,企业获取新用户的成本也越来越大。随着中国三大运营商竞争的加剧,产品和服务的同质化程度也越来越高,这使企业在老客户的保留上变得异常困难,客户流失率也在逐年上升,如何识别高风险流失客户并有效防止客户流失已经成为该行业管理者普遍关心的问题之一。
着眼于客户流失影响因素研究,运用社交网络分析方法,通过构造与网络结构相关的变量进行影响因素的探讨,运用逻辑回归方法构建客户流失预警模型。从社交网络的视角出发,利用客户的通话详单数据建立客户之间的通信网络,在自我网络的相关理论框架下,构建个体的度、联系强度、个体的信息熵3个自我网络特征变量。运用中国某移动运营商公司的月度客户数据(包括基础通信数据和通话详单数据),通过逻辑回归构建基于社交网络变量的客户流失预警模型。
研究结果表明,个体的度、联系的强度和个体的信息熵都对预测客户流失有显著效果。具体的,个体的度越大,联系强度越强,个体的信息熵越大,客户越不容易流失。外样本AUC值平均可以达到0.75以上,模型具有良好的预测精度。
研究结果对企业实践具有非常重要的意义,合作企业应用客户流失预警模型进行高风险流失客户的识别,预测精度可以达到70%,达到了企业的实践预期。客户流失预警模型可以帮助企业提前识别高风险流失客户,极大地降低企业维系客户的成本。建议企业管理者在未来更加关注与客户社交网络有关的变量,从网络结构的视角理解消费者行为,更好地进行客户关系管理。
英文摘要:
      In the recent years, it is observed that, given the prospering development of telecommunication industry, the companies experience costly investment in customer acquisition as the market becomes more saturated. Meanwhile in China, the fierce competition between three major operators has intensively promoted the homogeneity of products and services. This leads to the difficulty of retaining old customers in the sense that customer attrition is increasing annually. Therefore, the identification of and prevention for customer attrition are regarded as a key issue in the field of telecommunication management.
This paper focuses on the study of customer attrition. Inspired by the popularly used social network analysis method, we construct some network related variables to explore the influencing factors. A logistic regression is proposed to build customer attrition model. From the perspective of social network, we establish a customer communication network using their point-to-point communication data. Under the framework of ego network, this paper constructs three ego network featured variables, namely,degree, tie strength and ego entropy. The empirical data comes from one of the three major telecommunication companies in mainland China. These data includes communication bill and point-to-point communication data. A logistic regression is used to investigate customer attrition model based on these variables.
It is found that, the degree, tie strength, and ego entropy are all significant indicators in predicting customer attrition. Specifically, if a customer has a larger degree, a higher tie strength and a bigger ego entropy, then his attrition rate will be lower than others. The out of sample AUC value is about 0.75 on average, which reflects a relatively high prediction accuracy.
The results of this paper are of great importance to the practice of enterprises. The model has been adopted by the cooperative enterprise. They use the model to identify high risk customers who are going to leave, and the prediction accuracy can reach 70%, which meets the expectation of the enterprise. The proposed attrition model can help companies to identify their high-risk customers in advance. This greatly reduce the cost of maintaining existing customers. Through this study, we strongly recommend business managers should pay much attention to the social network-related variables of customers. It can help the company to better understand consumer behavior from the perspective of network structure and thus better customer relationship management.
查看全文  查看/发表评论  下载PDF阅读器
关闭