基于先验信息分类的客户消费动态终身价值估计研究

Customer Consumption Dynamic Lifetime Value Estimation Based on a Priori Information and Empirical Study

文章借助SMOTE方法平衡客户消费行为数据,结合受限玻尔兹曼机和区制转移动态因子模型,给出客户初步动态分类先验信息及处于各消费阶段的概率,在此基础上估计客户动态终身价值,并通过库尔勒香梨的网络销售样本数据进行实证分析。研究表明,SMOTE方法可以较好地增加客户的消费行为数据而不影响其统计特征;受限玻尔兹曼机可将库尔勒香梨的消费群体划分为统计特征完全不同的3类客户,区制转移动态因子模型估计得到的3类客户的动态特征有明显区别。另外,各类客户的动态终身价值均呈倒“U”形特征,与理论一致,仅消费一次客户在大多数时间点上终身价值会更高。

By employing SMOTE method to balance customer consumption behavior data and combining restricted Boltzmann machine with zone system transfer dynamic factor model,this paper gives the preliminary dynamic classification prior information of customers and the probability of each consumption stage.On this basis,it estimates customer dynamic lifetime val⁃ue and makes an empirical analysis of online sales sample data of Korla pears.The study shows that SMOTE method can better increase customers'consumption behavior data without affecting their statistical characteristics,the restricted Boltzmann ma⁃chine can divide consumer groups of Korla pears into three types with completely different statistical characteristics,and their dy⁃namic features are obviously different estimated by zone system transfer dynamic factor model.In addition,dynamic lifetime val⁃ue of customers of each type all shows"U"shape,which is in accordance with the theory.However,the lifetime value of those who only consumes once will be higher at most time nodes.