Hierarchical Clustering Algorithm For Customer Segmentation

Customer segmentation divides customers into several groups with different features, based on the unique features of different customer groups, enterprises formulate differentiated marketing strategy so that they can achieve efficient and accurate product promotion goals. Existing customer segmentation methods are mainly based on artificial rules or simple classification, which can be effectively applied to small-scale and low-dimensional data sets. However, effectiveness and correctness of these algorithms are greatly diminished when applied to the processing of large-scale and high-dimensional data sets. Customer segmentation based on clustering algorithm that can adapt to large-scale and high-dimensional customer data sets is proposed in this paper. The detailed contents are presented below:(1) To deal with the performance reduction of segmentation algorithm caused by high-dimension of customer data, attribute reduction algorithm that combines information entropy method and weighted PCA is proposed. Firstly, reduce the attribute quantitatively with the information entropy method; secondly, adopt attributes weighted reduction method eliminates the influence of irrelevant or weakly relevant attributes on the performance of the algorithm. The experimental results show that the proposed algorithm improves the clustering accuracy in the customer segmentation and public data sets compared with the weighted principal component analysis method, clustering accuracy rate increased by 1%-5%.(2) To solve low effectiveness of the segmentation algorithm caused by large-scale of customer data sets, grid density deviation sampling algorithm for customer segmentation is proposed. This algorithm divides each dimension's property with non-equal wide sliding window and merges similar windows so that the calculated quantity of the sampling algorithm will be reduced. The experiment proves that the proposed algorithm is more effective than the equal wide grid division method, in the case of the same clustering accuracy that the sampling ratio decreased by 1%-9%.(3) Aiming at the problem that the existing customer segmentation methods using only a single customer segmentation indicators for customers to be divided, can not meet the diverse needs of the marketing, hierarchical clustering algorithm for customer segmentation is proposed. This method divides the segmentation process into two steps: firstly, customers are divided into several preliminary customer settlements by the adaptive clustering algorithm; secondly, the demand for corporate marketing plans for multiple targets, implement spectral clustering algorithm and the quantitative weighted comprehensive evaluation indexes on the settlements formed to be divided that is good for marketing, so marketing departments to improve the efficiency of developing personalized services to all types of customers.(4) Based on the difference of contribution that the different types of customer attributes make when different clusters are divided, adaptive weight clustering algorithm for customer segmentation is proposed. Dynamically assigning weights to the customer attributes based on the degree of dispersion of data. Experiments show that this method can make the clustering results closer to the actual center of the customer, the correct rate of clustering in public data is improved by 4%-6%.Based on large-scale and high-dimensional customer segmentation scheme for customer segmentation is proposed in this paper. Experimental results show that the method can effectively improve the accuracy of customer segmentation, in order to provide reliable algorithm for the marketing departments to automatic accurate and effective promotion of products to customers.