基于评论行为的客户终身价值模型改进

发布时间：2018-06-27 16:13

本文选题：RFM模型 + 客户流失行为　；参考：《北京邮电大学》2015年硕士论文

【摘要】：客户终身价值一直以来都是企业进行客户管理以及相关营销活动的重要依据。通过对群体层面的客户终身价值和个体层面的客户终身价值进行分析,企业能够对现存客户的价值进行识别,并且不断改进客户管理方向以实现客户价值最大化。本文主要从群体层面和个体层面两个维度来进行客户终身价值的计算,分别通过两种不同的数学模型来对客户的价值进行全方位的分析,以期为企业对客户的管理决策奠定基础。在Web2.0时代越来越多的消费者乐于在购物网站、点评类网站以及社交类网站上以文字的形式,发布对相关产品、服务、品牌或企业的评价。众所周知,对于企业而言,用户每产生一次购买行为都会为企业带来一定的利润,会给企业带来一定的直接价值,而用户每产生一次评论行为是否同样会为企业带来一定的影响与价值。现阶段研究者仅从消费者的购买行为出发对客户的终身价值进行建模,并未有研究者综合考虑了客户的购买行为和评论行为。由于该研究点较为新颖,未出现相关的研究,因此本文决定将评论行为引入客户终身价值建模领域,来进行相关的尝试和探索。在研究大众点评网客户的过程中,本文发现同一消费者会同时使用其点评平台和团购平台,从而使得用户自身的两种行为会产生相互影响,因此针对这种客户特征,本文尝试建立了群体层面和个体层面两种改进的客户终身价值模型。本文的模型主要由两部分组分：群体层面的客户终身价值模型和个体层面的客户终身价值模型。针对群体层面的模型研究,本文选取RFM作为基础模型,将评论行为引入模型中,提出了综合考虑两种行为的改进RFM模型；针对个体层面的模型研究,本文首先利用Logit回归模型对评论行为对购买行为的影响进行研究,提取出关键变量,并将这些变量作为协变量引入Pareto/NBD模型中,进一步的改进了模型对购买金额进行预测;再次向件利用Logit模型和改进Pareto/NBD的方法进行评论次数的预测；最后提出针对个体层面的客户终身价值模型。实证结果显示,预测模型都取得了很好的拟合和预测效果。本研究发现了消费者自身评论行为对于其购买行为的影响,并且通过实证分析达到了较好的预测结果,这对于企业今后进行真正高价值用户的识别,改进相关管理措施提供了重要依据。
[Abstract]:Customer lifetime value has always been an important basis for enterprise customer management and related marketing activities. By analyzing the customer lifetime value at the group level and the customer lifetime value at the individual level, the enterprise can identify the existing customer value and continuously improve the direction of customer management to maximize customer value. This article mainly from the group level and the individual level two dimensions carries on the customer lifetime value computation, respectively through two kinds of different mathematics models carries on the omni-directional analysis to the customer value. In order to establish the foundation for the enterprise to the customer management decision. In the Web 2.0 era, more and more consumers are willing to publish reviews of related products, services, brands or enterprises in the form of text on shopping sites, comment sites and social networks. As we all know, as far as enterprises are concerned, every time a user produces a purchase behavior, it will bring a certain profit to the enterprise, and will bring a certain direct value to the enterprise. And each time the user produces a comment behavior will also bring certain impact and value for the enterprise. At present, the researchers only model the customer's lifetime value based on the consumer's purchase behavior, and no researchers have considered the customer's purchase behavior and comment behavior synthetically. Because the research point is relatively novel and there is no related research, this paper decides to introduce the comment behavior into the field of customer lifetime value modeling to try and explore it. In the process of studying Dianping customers, this paper finds that the same consumer will use both the comment platform and the group purchase platform, so that the two behaviors of the users themselves will have mutual influence, so this paper aims at the characteristics of this kind of customers. This paper attempts to establish two improved customer lifetime value models: group level and individual level. The model is divided into two parts: customer lifetime value model at group level and customer lifetime value model at individual level. In this paper, we select RFM as the basic model, introduce the comment behavior into the model, put forward the improved RFM model which considers two kinds of behavior synthetically, and study the model of individual level. In this paper, logit regression model is first used to study the effect of comment behavior on purchase behavior, and key variables are extracted, and these variables are introduced into Pareto / NBD model as covariables to further improve the model to predict the purchase amount. At last, the author uses logit model and improved Pareto / NBD method to predict the number of reviews. Finally, a customer lifetime value model for individual level is proposed. The empirical results show that the prediction models have achieved good fitting and prediction results. This study finds out the influence of consumers' own comment behavior on their purchasing behavior, and achieves a good prediction result through empirical analysis, which is useful for enterprises to identify real high-value users in the future. To improve the relevant management measures to provide an important basis.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F274

【参考文献】