Segmenting the e-commerce market using the Generative Topographic Mapping

Abstract

The neural network-based Generative Topographic Mapping (GTM) (Bishop et al. 1998a, 1998b) is a statistically sound alternative to the wellknown Self Organizing Map (Kohonen 1982, 1995). In this paper we propose the GTM as a principled model for cluster-based market segmentation and data visualization. It has the capability to define, using Bayes' theorem, a posterior probability of cluster/segment membership for each individual in the data sample. This, in turn, enables the GTM to be used to perform segmentation to different levels of detail or granularity, encompassing aggregate segmentation and one-to-one micro-segmentation. The definition of that posterior probability also makes the GTM a tool for fuzzy clustering/segmentation. The capabilities of the model are illustrated by a segmentation case study using real-world data of Internet users opinions on business-to-consumer electronic commerce © Springer-Verlag Berlin Heidelberg 2000.