基于改进的Pareto/NBD模型预测博客用户在线行为  

Forecasting Blog Users'Online Behavior Based on Improved Pareto/NBD Model

作　　者：蔡瑞[1] 齐佳音[1] 

机构地区：[1]北京邮电大学经济管理学院,北京100876

出　　处：《统计与信息论坛》2013年第6期69-75,共7页Journal of Statistics and Information

基　　金：国家重点基础研究发展计划<面向服务的未来互联网体系结构与机制研究>(2012CB315805);国家自然科学基金项目<半契约情景下的客户终生价值建模及营销决策研究>(71171023);北京市自然科学基金项目<基于微博的非常规突发事件信息沟通决策研究>(9122018);教育部新世纪优秀人才支持计划<组织客户的价值管理理论及决策支持研究>(NCET-10-0241)

摘　　要：博客用户在线行为分为发文行为和流失行为。由于这两种行为分别与交易过程中客户的购买行为和流失行为具有相似性,选择借鉴客户基分析中的Pareto/NBD模型进行预测。考虑到用户间交互性对博客用户在线行为具有重要影响,通过比例风险模型向经典的Pareto/NBD模型中加入体现用户间交互性的协变量。Pareto/NBD模型经过改进,实现了对博客用户在线行为的预测。实证研究以用户博客空间中的总评论量和总浏览量作为协变量。数据分析结果显示,当使用总评论量作为影响流失行为的协变量时,改进模型的预测精度显著提高。进一步分析还发现,总评论量对博客用户"存活"时长的正向激励存在着阈值。The blog users＇ online behavior was divided into publishing behavior and churn behavior. Since these two behaviors are similar to customer buying behavior and churn behavior in the transaction process respectively, this paper chose to learn from the Pareto/NBD model in the field of customer base analysis to forecast them. Considering that the interaction between blog users has a great influence on their online behavior, this paper incorporated covariates which can reflect such interaction into the classical Pareto/NBD model through the proportional hazards model. The improved Pareto/NBD model realized the blog users online behavior forecasting. The empirical study chose both the total amount of comments and page views in the user＇ blog space as covariates. The results show, when using the total amount of comments as a covariate to the churn behavior, the improved model has significantly improved the forecasting accuracy. Further analysis also finds that the positive impact of the total amount of comments on blog users＇ survival duration has a threshold.

关 键 词：客户基分析Pareto NBD模型 比例风险模型 在线行为 博客用户行为 

分 类 号：O211.67[理学—概率论与数理统计]