一种针对不完全观测数据的消费者聚类方法

摘要

现有聚类方法都是基于消费者全部的行为信息，对于观测不完全的信息，提出了三阶段聚类方法。首先，使用样本数据的全部信息对消费者聚类；接着仅使用人口统计变量建立分类模型；最后对上述结果进行修正。三阶段聚类方法最大优点是可以将没有入选样本的个体分配到由样本个体得到的行为集群中去，将这个方法应用于电视行业，得到了很有实际应有价值的结果。%Most of existing clustering methods are based on consumers'full information ,so they can't be applied to incomplete data . T his article proposes a new three‐step segmentation method to solve this problem .To be specific ,in first step ,sample consumers are grouped into different clusters with full information ;in second step ,classification model is built with only demographic information ;in third step , the model is corrected .The key advantage of the new method is that it can classify a new sample with little information into a particular cluster . Last , the method is applied to television audience segmentation problem .The results are interesting and useful .