Title: A Bayesian spatial market segmentation method using Dirichlet process-Gaussian mixture models
Authors: Won Chang - University of Cincinnati (United States) [presenting]
Sunghoon Kim - Arizona State University (United States)
Heewon Chae - Arizona State University (United States)
Abstract: A new spatial market segmentation method is proposed by using a Bayesian spatial Dirichlet process-Gaussian mixture (SDPGM) model. The approach forms segments of spatial locations that show a similar relationship between service factors and customer satisfaction. Our method employs a SDPGM model in the spatial domain and hence automatically determines the number of segments and the membership of spatial locations simultaneously in a unified framework. We also incorporate ridge and lasso regularization in parameter estimation to better select statistically significant service factors, which is important for efficient resource allocation in marketing. Our simulation study confirms that (i) the proposed approach can successfully identify the hidden spatial segmentation structure only based on the spatial distribution and the predictor-response relationship and (ii) the ridge estimator shows a better performance in identifying non-significant variables and hence leads to a better identification of key market drivers than lasso. We apply the proposed approach to an online customer satisfaction data set for the restaurants in Washington DC area collected by Yelp. The results provide an interesting insight on the key drivers of customer satisfaction in different sub-regions of the area.