Research And Implementation On Customer Segmentation Model Based On Clustering

Along with the enterprise focusing on customers rather than products, how to make use of clustering technology to segment customers properly is the most important problem for the business circle and academe. Due to the drastic competition, both the clustering technology and the valuation of customers are required to be more and more effective in customer segmentation system. However, these problems have not been solved yet. This paper mainly encircles these two aspects, targeting at implementing a more effective model to segment customers.Firstly, K-means is sensitive to the selection of initial centers, so we adopt a neighborhood density method——NK-means for effectively selecting initial cluster centers in K-means clustering. This improved algorithm clusters more quickly and correctly than K-means. Experimental results have shown its significant improvements in clustering accuracy in comparison with the random K-means and the refinement K-means. These results also show that the running time of NK-means linearly increased with the number of points and the number of dimensions.Secondly, we find that there exist some problems when we employ K-means to segment customers attribute to its clustering standard which bases on distance. In order to describe the distance between customers rightly, we propose a purchase-based similarity measure. It provides more dynamic similarity between customers than traditional ones and achieves more accurate segmentation of customers.Thirdly, the traditional CLV models considered the socio-demographic and the behavioral factors except the network influence between customers. So we propose a model based on social influence between customers. It not only combines customers'socio-demographic and behavior, but also considers the network influence between them. We evaluate customers'value through its four components: basic value, potential value, network value and retention rate. Meanwhile we use RFM model to assist this model.