Abstract:

Customers are the decision makers in the satisfaction measurement for a product. It is significant to identify the heterogeneous features of different customer segmentations through customer classification. The primary cause of the diversity and conflict of customer evaluation characteristics is the heterogeneity of customers as decision makers. The heterogeneity of customers derives from the characteristics of customers, which includes categorical and numerical attributes simultaneously. A classification method of customers based on the punitive competition mechanism is proposed. Firstly, the initial clustering centers are determined according to the distribution regularities of the numerical and categorical attributes. Secondly, a unified similarity measurement is established and the punitive competition mechanism is introduced to realize the iteration and the auto-optimization of the number of clusters. Finally, the heterogeneous customer classification of a specific product is implemented to expound the feasibility of the proposed method. To elaborate the effectiveness of the proposed approach, the standard dataset “Heart Disease” is used to establish the comparison analysis with two classical clustering approaches K-means, and K-prototypes.

Key words: classification of decision makers, customer classification, clustering for mix-valued dataset, punitive competition mechanism

摘要：

顾客作为产品满意度测度过程中评价决策的主体，对其进行分类研究，识别不同顾客群体异质评价特征具有重要意义。顾客评价特征存在多元性和冲突性，根本原因是顾客作为决策者的异质性，而顾客的异质性来源于顾客本身属性，包含分类型属性和数值型属性。提出了一种基于惩罚竞争机制的混合属性顾客分类方法，根据数值型和分类型属性值的分布规律，给出了混合数据初始聚类中心的确定方法；建立了统一相似性度量模型，并引入惩罚竞争机制，实现了聚类过程中的基本迭代和自动优化聚类数。以某产品异质顾客分类问题为例验证了所提方法的可行性，继而通过“Heart Disease”标准数据集将所提算法与K-means和K-prototypes两种经典聚类算法进行对比，验证了该方法的有效性。

关键词: 决策群体分类, 顾客分类, 混合数据聚类, 惩罚竞争机制