Determining a retail basket’s purpose is a key part of driving understanding of shopping behavior. One way that dunnhumby has attempted to build understanding in the past is through topic modelling, where the basket is reimagined as a textual document being examined. Building off those foundations, the definition of a document is shifted to incorporate multiple baskets for identifiable customers. Using transaction data from a North American retailer, dunnhumby trained a topic model on millions of customers to learn several concepts that reflect individualized shopping behavior. The topics found correspond with distinct behavioral patterns and provide a way to tag customers with topics describing their shopping behavior.
Supported by
