Product Recommendations with Multi-Armed Bandit

摘要

1.Intelligent retailing. 2.Multi-Armed Bandit with Thompson Sampling. 3.Retail Product Recommendations. 4.Next Steps. 5.Summary. 6.Q&A. INTELLIGENT RETAILING. Achieve revenue maximization and provide differentiated experiences. Traveler expectations are changing. Increase revenue while enhancing guest experience with solutions that span full journey Travel industry is laser-focused on becoming better retailers. Changes in the market Evolving customer behaviors Emerging technologies. Relevant,personalized offers. Ready to intelligently retail? What are the needs of the customer? What intelligent infrastructure is needed to meet the company needs? How to provide the best retailing options and win against competition? MULTI-ARMED BANDIT WITH THOMPSON SAMPLING. Exploration vs Exploitation. An intelligent platform for retailing 1.Select an ordered list of products to display (MAB-TS). 2.Infinite SKUs - anything store. 2.1 things (robe, blanket,…). 2.2 policies (early check in, late check out,…). 2.3 services (airport transport, flower delivery,…). 2.4 experiences (massage, bike rental, …). 3.SKUs are not attached to the room. Stochastic Multi Armed Bandit Problem 1.A slot machine with multiple arms. 2.Each arm produces an i.i.d. random reward independent of the past and other arms. 3.Parameters of the reward distribution are fixed but unknown. 4.A reward can be observed only after playing an arm.