Anonymous Shoppers: Zero-Shot Associative Learning for Purchase Predictions

Anonymous Shoppers: Zero-Shot Associative Learning for Purchase Predictions

Abstract

Addressing the challenge of predicting purchases of first-time or anonymous customers, this research
introduces a novel zero-shot learning model adept at making accurate purchase predictions without any prior history of the shopper. Our model operates on the sole data point available—the customer’s current shopping basket—amidst expansive and diverse product assortments typical of modern retail environments. We draw on marketing theory, particularly product associations and purchase complementarity, to understand and model the contents of a shopper's basket. By delving into the composition of shopping baskets, we emphasize the significance of learning how products complement each other in real-time purchasing scenarios. Central to our approach is its capacity to separate meaningful purchase patterns from inherently noisy market basket data. Through extensive simulation and empirical application, we contrast our model against alternative methods and reveal key determinants of purchase prediction accuracy. We further investigate the extent to which purchase predictions for anonymous shoppers depend on meaningful add-to-basket sequences. Our work contributes new marketing insights to the recommender literature and provides actionable guidelines for retailers on tailoring product recommendation systems to their specific retail settings.

More information on Dr. Daniel M. Ringel can be found here.

The talk will take place from 12:15-13:30 in RuW 1.201 and via Zoom.