基于健康度的客户分群及流失预警算法研究  

Research on algorithms for customer grouping and loss warning based on health degree

作　　者：曾国文[1] 钟玲 罗琼华 ZENG Guo-wen;ZHONG Ling;LUO Qiong-hua(China Mobile Group Guangdong Co.,Ltd.Dongguan Branch,Dongguan 523129,China)

机构地区：[1]中国移动通信集团广东有限公司东莞分公司,东莞523129

出　　处：《电信工程技术与标准化》2023年第11期49-55,共7页Telecom Engineering Technics and Standardization

摘　　要：为了能更好地开展移动通信客户分群运营和客户流失预警,本文从多个维度分析客户业务使用“不健康”的复杂因素,引入标准化和熵值法重组指标变量,采用基于标准分的分段排名算法,计算客户的长期趋势、短期趋势、现状水平和弹性空间,预测客户离网概率并据此进行段内二次排序,实现客户“健康状态”的精准定位,最终根据“健康程度”细分客群,并开展分群施策,向“健康客户”要收入,向“高危客户”争保有。通过历史数据证明,本算法识别的“高危群体”离网率高达44.7%,并集中分布在长期趋势为下降型的客群里,证明了算法的准确性。In order to better carry out mobile communication customer clustering operation and customer churn warning,this article analyzing the complex factors of unhealthy customer business usage from multiple dimensions.Introducing standardization and entropy method to restructure indicator variables,adopting a segmented ranking algorithm based on standard scores,calculate customers'long-term trends,short-term trends,current levels,and elastic space,predict the probability of customer churn and perform secondary sorting within the segment based on it,accurate positioning of customer“health status”.Final segmentation of customer groups based on"health level",and carry out group based policy implementation,seeking income from"healthy customers"and competing for retention from"high-risk customer".Prove through historical data,the dropout rate of the"high-risk custome"identifi ed by this algorithm is as high as 44.7%,and concentrated in customer groups with a long-term downward trend,proved the accuracy of the algorithm.

关 键 词：健康度 精准分群 流失预警 智能算法 

分 类 号：TN929.5[电子电信—通信与信息系统]