New data mining and marketing approaches for customer segmentation and promotion planning on the Internet

Various business problems are of interest to both data mining and marketing academic communities, such as market segmentation, direct marketing, targeted marketing, personalization/customization, cross selling and discovering customer lifetime value. Yet, these two disciplines have very different approaches to analyzing these problems. Data mining research incorporates methodologies from various research disciplines such as statistics, machine learning, database technology, optimization and pattern recognition, and hence has a richer pool of knowledge/model representation. It focuses more on the effectiveness of problem solving, and pays greater attention to the actual performance on data. In contrast, marketing research advocates more theory-based analysis and its theories are often built upon statistics, economics, econometrics and other social sciences. The broader research goal that motivates the dissertation is to integrate and compare methods from marketing research and data mining research for solving marketing problems. This dissertation consists of two essays that contribute to this goal. In each of the two essays in this dissertation, both data mining and marketing approaches are used in addressing selected marketing problems. The first essay develops new data mining theories and applies them to market segmentation problems. At the same time, comparisons are made between proposed methods and commonly used methods from both data mining and marketing research, and the results show that the proposed methods perform well. The second essay focuses on a specific marketing problem: the relationship between free shipping promotions and Internet shopping behavior. It uses original analytical models and empirical methods from both marketing and data mining research to address the problem. Each essay is independently evaluated on different data sets. These data sets all originate from the Web, since the marketing research community has not developed a significant body of research based on Internet data, due partially to the overwhelming amount of data, noise in the data and the extent of effort needed to preprocess the data.