earticle

영어

In this paper we consider utilizing dynamic Bayes network methods to model different types of long-term market basket analysis problems. The proposed approach can be used for learning and inference objectives, for both maximum likelihood parameters of a model given the structure and the dynamics of the structure itself. We use Markov Chain Monte Carlo (MCMC) methods such as Metropolis-Hastings to approximate high order integrals of the joint probability distributions and use results from the dynamic Bayes network literature to devise learning algorithms in a time-series market basket data setting. We illustrate the implementation of the proposed approach with real world data on the joint association structure of low-dimensional models and show that there are clear differences in long-term promotion effects depending on the nature of product categories and confirm the instantaneous, lagged effect of promotion activities and also the recency aspect of consumer choice behavior in multiple product category setting. The findings of our paper help further the understanding of consumer behavior through the dynamic analysis of market baskets and provide managerial insights on the use of big data and promotion strategies in offline and online retail stores.