This study proposes a new collaborative filtering model that integrates Restricted Boltzmann Machines. The proposed two-stage model is applied to household-level supermarket purchase data. Results show that our model fits the data better and outperforms existing collaborative filtering methods in predicting shopping patterns. The proposed model also improves interpretations of market complexity and common causes of coincidence associated with customers’ multi-category purchases.

• Myounggu Lee(South Korea, KAIST)
• Jihoon Cho(USA, Kansas State University)
• Youngju Kim(France, Neoma Business School)
• Hye-Jin Kim(South Korea, KAIST)