An Integrated Approach For Market Segmentation And Visualization Based On Consumers' Preference Data

yu xiang lv,gang guo,daijie cheng

DOI: https://doi.org/10.1109/icmlc.2004.1382050

2004-01-01

Abstract:The research in market segmentation includes two main parts. We first focus on discussing the market segmentation problem by applying clustering technique in data mining discipline. The partition of market is based on users' preference data and not on the commonly used one, users' attribute data. In that way, the definition of distance between two customers by their preference to a set of specified competing products is given. In stead of starting from scratch, the self-organization feature map is adopted as a basic clustering framework. In order to process the preference data, some necessary modifications are made. Both theoretic analysis and practical experiment are presented in this paper, which make us confident of that the algorithm we proposed has excellent performance and could discover the potential clustering patterns in the complex datasets. The second part is focus on displaying market segmentation structure. We apply visualization technique to representing the market structure clearly in a two dimensional plane so that the marketers can make their market strategies easier. The two main parts are organized as an integrated approach. Such an approach includes three core steps: preference data collecting step, preference data clustering step, by SOM neural networks and visualization step by ideal point model. There are three main advantages of the approach: firstly, the approach is based on well-defined mathematic models and can be supported by a series of numeral methods. Secondly, it does not have to face the tough market variable selection problem because we focus on preference data, not on evaluators' attribute data (demographic or geographic data etc.). Finally, the approach can produce multi-scale view of market segmentation results. The experiments show that the approach yields meaningful results and is comparable and complemented to the most general ones.