基于流形学习的客户价值分析研究

Study on the Customer Value Analysis Based on Manifold Learning

为了解决数据高维、海量导致聚类算法处理效果不佳的问题,提出将流形学习理论引入客户关系管理进行聚类研究。为了较好的分析客户价值,在Kmeans聚类的基础上引入流形学习理论。客户价值分析一般包含数据的抽取、探索以及预处理、模型建立几个步骤。在模型建立过程中一般采用Kmeans聚类实现。使用流形学习的谱聚类来替代Kmeans聚类。使用泰迪杯数据挖掘大赛中的数据进行试验,通过实验的雷达图可以看出,谱聚类与Kmeans聚类具有相似的分类构成。同时对于分类后的数据进行规约并绘制散点图,比较后发现,谱聚类后的数据类间相似度比Kmeans高,表明将流形学习方法引入客户价值分析,对于聚类稳定性有一定改善。

In order to solve the problem of the present data high dimension and the large amount of the current clustering algorithm,it is proposed to introduce the theory of manifold learning into customer relationship management.In order to better analyze the value of the customer,the theory of flow-based learning was introduced on the basis of the Kmeans clustering.Customer value analysis generally includes extraction of data,exploration of data,and several steps of preprocessing and modeling.In the process of modeling,Kmeans clustering is generally implemented.The spectral clustering of manifolds is used to replace the Kmeans clustering.In this paper,the data in the data mining contest of the teddy cup are used to experiment.According to the radar map of the experiment,it can be seen that the spectral clustering and Kmeans clustering have similar classification.In addition,the data of the classification after classification are also plotted and the scatter plots are drawn.After comparison,the similarity between the data classes after the spectral clustering is higher than that of Kmeans.Finally,it is concluded that the method of manifold learning can improve the stability of the cluster in the customer value analysis.