Abstract: Aiming at the customer segmentation problem of data intensive enterprises,a novel customer segmentation framework based on selective clustering ensemble was developed.In clustering generation step,a unified customer view was constructed based on data sources and business requirements and the customer attributes were split into various subsets,on which a clustering algorithm was run to label the customers with clustering labels.A new evaluation function was introduced to select the most qualified label vectors for clustering ensemble.In clustering ensemble step,an overlap matrix recorded the number of customers that share the same labels was built,and the component clusters for ensemble were selected with the help of overlapping weight.Empirical results on real world customer data demonstrated that the proposed framework could effectively identify customer groups with different values and consumption habits,and thus provided supporting details for business and marketing decision.

Key words: customer segmentation, clustering, ensemble learning, k-means algorithm

摘要： 针对数据密集型企业的客户细分问题,提出一种基于选择性聚类集成的客户细分框架。在聚类集体生成阶段,根据数据来源和业务需求构建统一的客户视图,将客户特征划分为若干子集后再分别对客户对象聚类,通过评价函数选择高质量的个体标记向量生成聚类集体;在聚类集成阶段,构建记录簇标记所覆盖的相同对象个数的重叠矩阵,利用重叠矩阵计算各簇权值,最后选择最具代表性的簇参与集成。通过某企业客户细分的实证研究表明,该框架可以有效识别出不同价值和消费行为习惯的客户群,为企业制定产品营销方案提供依据。

关键词: 客户细分, 聚类, 集成学习, k均值算法