A product segmentation approach and its relationship to customer segmentation approaches and recommendation system approaches
MetadataShow full item record
As part of their customer management strategy, retailers with large, multi-category offerings need to present their products in ways that help target customers search and choose from those offerings. In the first essay of this dissertation, a product segmentation approach is proposed. The proposed approach gives retailers a methodology for directly identifying customer-centric, cross-category, product segments from large numbers of products in multiple categories such that products within a segment are purchased by the same type of customers. In addition, the research examines the relationship between the proposed product segmentation approach and a parallel customer segmentation approach. The close relationship between the approaches suggests that the segments of products and customers inferred from each approach will be equivalent. However, the results show that this is not the case because of the aggregation constraint imposed on customers in the product segmentation approach and on products in the customer segmentation approach. Further, the results indicate that the product segmentation approach provides better recommendations of products for a customer to purchase, while the customer segmentation approach provides better recommendations of customers for a product to target. To increase customer repurchasing and loyalty, retailers with large offerings are increasingly employing recommendation systems. The second essay of this dissertation contributes to our understanding of recommendation systems in three respects. First, we present a new methodology, attribute-based co-clustering, which incorporates customer characteristics and product attributes to produce recommendations. This approach has not previously been evaluated in recommendation system contexts. Second, we compare the performance of the proposed approach with a related latent class segmentation approach and a widely applied collaborative filtering approach. Third, we identify factors that impact recommendation quality in two contexts: recommending a set of products for a customer to purchase and recommending a set of customers for a product to target. Results indicate that latent class segmentation quality improves in databases with large samples and strong predictors that characterize customers’ preferences, while collaborative filtering quality improves with greater data density. Attributebased co-clustering quality improves when customer and product attributes are predictive of choices, however, it is more stable with respect to data distribution.