The generative topographic mapping as a principal model for data visualization and market segmentation: an electronic commerce case

The process of extracting knowledge from data involves the discovery of patterns of interest which may be implicit, for instance, in speciÞc clusters of data points. In the context of Internet retailing, Þnding clusters of typical consumer types is among the most important uses of data mining techniques. Cluster-based market segmentation models, grounded on surveys of customer opinion, can give the online retailer a competitive edge, forming the basis for effective targeting and enabling the redirection of made-to-measure content towards the customer. The Generative Topographic Mapping (GTM) is proposed as a statistically principled technique for cluster-based market segmentation. In this non-linear latent variable model, a posterior probability of cluster membership can be deÞned for each individual, providing a robust framework for the visualization of high dimensional data and the segmentation to different levels of granularity. The advantages of the GTM over the well-known Self-Organizing Map (SOM), to which it is an alternative, are described and this new model is applied in a business-to-consumer e-commerce case study. In addition, an entropy-based measure is deÞned to quantify the information content of the GTM unsupervised maps about an externally imposed class label.