基于共享最近邻的客户交易数据聚类算法

A Customer Transaction Data Clustering Algorithm Based on Shared Nearest Neighbors

利用客户交易数据聚类分析,可得到更优异的客户细分效果,有助于企业更详实地了解消费者,制定精准的营销策略。PurTreeClust是一种新型的客户交易数据聚类算法,定义了一种新型的度量方式PurTree距离,可以很好地分析处理具有层次树结构的交易数据,但未考虑近邻点的影响,仅将交易树分配到距离最近的聚类中心所属类簇,容易出现错误的交易树分配。该文利用交易树之间的共享最近邻信息,提出一种客户交易数据聚类算法。该算法在聚类分配时,充分利用共享最近邻,首先分配类簇的从属交易树,然后分配类簇的可能从属交易树,实现聚类分配,可发现更加紧凑清晰的类簇,并避免了交易树错误分配,改善了客户细分效果。最后采用6个真实客户交易数据集进行实验,验证了该算法的有效性。

By clustering analysis of customer transaction data,better customer segmentation effect can be obtained,which is helpful for enterprises to have a more detailed understanding of consumers and develop accurate marketing strategies.As a new clustering algorithm for customer transaction data,PurTreeClust defines a new measurement method,PurTree distance,which can analyze and process transaction data with hierarchical tree structure.However,without considering the influence of neighboring points,only the purchase tree is allocated to the class cluster belonging to the nearest cluster center,so the wrong purchase tree allocation is prone to occur.We propose a clustering algorithm for customer transaction data using the shared nearest neighbors information among purchase trees.The algorithm makes full use of the shared nearest neighbors to achieve cluster allocation.Firstly,the subordinate purchase tree of the cluster is allocated,and then the possible subordinate purchase tree of the cluster is allocated to realize cluster allocation.It can find more compact and clear clusters,avoid the wrong allocation of the purchase tree,and improve the effect of customer segmentation.Finally,experiments on six real customer transaction datasets verify that the proposed algorithm is more effective.