Abstract: In view of the fact that the airlines evaluate the clustering effect in the clustering of customers and determine the best k-valued contour coefficient method, the time complexity is too high (O2) and the accuracy is low, firstly, the paper proposes an improved contour coefficient method, which uses the distance calculation between objects and clusters or different cluster centers to replace the distance calculation between similar or heterogeneous objects, and determines the contour coefficient adjustment position and weight adjustment by the clustering effect change rate; secondly, the paper builds a clustering model based on real-time airline customer data after pre-processing and feature selection, and determines the optimal customer grouping by means of improved contour coefficient method, and then constructs user profiles; finally, the paper describes the characteristics of different airline group customers and proposes corresponding personalized service measures to assist airlines to provide personalized products and services to customers. The results show that the accuracy of the improved contour coefficient method under different sample sizes is significantly improved and the operating efficiency is significantly improved and the airline customer grouping results, which are based on the improved contour coefficient method, are in line with objective reality, and the proposed service measures can provide reference for airlines to maximize customer demands and increase customer satisfaction.

Key words: improved silhouette coefficient, the optimal k value, customer clustering, feature selection, user profile

摘要： 鉴于航空公司在客户聚类分群中对聚类效果进行评价并确定最佳k值的轮廓系数法存在时间复杂度过高O(n2)以及准确率较低问题,文章首先采用对象与同簇或不同簇中心间距离计算来替换同类或异类对象间的距离计算,并通过聚类效果变化率确定轮廓系数调节位置及调节权重,提出一种改进的轮廓系数法;其次,基于预处理且特征选择后真实航空公司客户数据构建聚类模型,借助改进轮廓系数法确定最优客户分群,并构建用户画像;最后,针对不同航空公司分群客户进行特征描述并提出相应个性化服务措施,辅助航空公司为客户提供个性化产品与服务。实证研究结果表明:不同样本量下改进轮廓系数法的精确率和运行效率均有所提升;基于改进轮廓系数法的航空公司客户分群结果符合客观实际,所提服务措施为航空公司最大化客户需求、提高客户满意度提供借鉴。

关键词: 改进轮廓系数, 最佳k值, 客户分群, 特征选择, 用户画像