Bayesian Co-Clustering of Consumer Data

• Activity Number: 403
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): Ewa Nowakowska*+ and Joseph Retzer
• Companies: GfK and CMI Research
• Keywords: segmentation ; bayesian ; co-clustering ; dyadic data

Abstract

We present an overview of Bayesian co-clustering, which is a specific segmenting approach that allows for simultaneous partitioning of rows (respondents) and columns (attributes) of a data matrix. This is a tool for analysis of dyadic data, where observations are made of pairs of row and column objects. The method is able to recognize and utilize such inherent dyadic relationships, which helps accurately determine the data structure but is typically disregarded by classical approaches. The resulting row and column clusters (co-clusters) can be further profiled on covariates to facilitate their understanding and possible targeting. We show how the method can be applied for customer-product relation data, which normally presents a segmenting challenge due to its binary and frequently sparse structure. Also, Bayesian co-clustering allows for mixed membership, which is an additional advantage and a natural expectation in the context where each customer is likely to be in a different relationship with different products.