Paper Type
short
Description
Customer data today typically include a large number of customers with disaggregated information about the individual customers’ behaviors. There is a substantial degree of heterogeneity across individual customers in a customer dataset, but often not enough information about individual customers to infer about the individuals’ preferences and to predict the individuals’ decisions. We propose an approach for organizations to acquire individual-level data for constructing hierarchical Bayesian models for customer analytics applications. The proposed approach adapts a Kullback-Leibler divergence criterion to measure the value of additionally acquired information regarding the quality of estimation and prediction with hierarchical Bayesian models. Markov chain Monte Carlo simulation method is used for estimating the Bayesian model parameters. We develop a set of algorithms to select additional data such that the value of the acquired data is optimized. The results of a preliminary experiment demonstrate the effectiveness of our approach.
Recommended Citation
Liu, Xiaoping and Li, Xiaobai, "Acquiring Heterogeneous Customer Data for Business Analytics" (2019). ICIS 2019 Proceedings. 30.
https://aisel.aisnet.org/icis2019/data_science/data_science/30
Acquiring Heterogeneous Customer Data for Business Analytics
Customer data today typically include a large number of customers with disaggregated information about the individual customers’ behaviors. There is a substantial degree of heterogeneity across individual customers in a customer dataset, but often not enough information about individual customers to infer about the individuals’ preferences and to predict the individuals’ decisions. We propose an approach for organizations to acquire individual-level data for constructing hierarchical Bayesian models for customer analytics applications. The proposed approach adapts a Kullback-Leibler divergence criterion to measure the value of additionally acquired information regarding the quality of estimation and prediction with hierarchical Bayesian models. Markov chain Monte Carlo simulation method is used for estimating the Bayesian model parameters. We develop a set of algorithms to select additional data such that the value of the acquired data is optimized. The results of a preliminary experiment demonstrate the effectiveness of our approach.