| Chapter 2: | Segmenting Customer Transactions Using a Pattern-Based Clustering Approach |
distinct home pages and recommends new products for customers based on personalization models built from data. Most credit card and cellular phone fraud alerts are also issued based on analysis of customer-level data (Fawcett and Provost 1997). Consumer brand choice models and pricing models are heavily used in marketing endeavors (Bell and Lattin 2000, Danaher et al. 2003).
While the expectation for customer-level data analysis is high, there are still problems with existing analytical methods. For example, consumers still receive a significant amount of direct mail advertising products for which they have no interest; online recommendations are still far from perfect (Riedl 2001). In order to create more successful personalized systems and build more accurate consumer behavior models, firms must understand their customers better by collecting more information and better analyzing transaction data. There has been much research in this direction, and clustering transactions to discover segments has been one research stream that has generated a variety of useful approaches (Strehl and Ghosh 2000).
In the marketing literature, market segmentation approaches have often been used to divide customers into groups in order to implement different strategies (Hofstede 2002). It has been long established that customers demonstrate heterogeneity in their product preferences and buying behaviors (Allenby and Rossi 1999) and that the model built on the market in aggregate is often less efficient than models built for individual segments. Much of this research focuses on examining how variables such as demographics and socioeconomic status can be used to predict differences in consumption and brand loyalty. Distance-based clustering techniques, such as k-means (Hartigan 1975), and parametric mixture models, such as Gaussian mixture models (Fraley and Raftery 1998), are two main approaches used in segmentation. While both of these approaches have produced good results in various applications, there are well-known drawbacks to these methods (Wedel and Kamakura 1998). While distance-based clustering in an n dimensional space