Learning Market Dynamics for Optimal Pricing

on YouTube ↗34m

" summary

Airbnb clusters millions of unique listings into demand-driven groups using listing embeddings from view sessions and a recursive partitioning tree, then models per-cluster booking arrivals as a gamma-shaped lead-time distribution plus Fourier cycles, predicting five parameters (alpha, beta, omega, rho, phi). This cuts prediction scale from a potential trillion per day to roughly a thousand-fold smaller problem, feeds Smart Pricing, improves fit to empirical cumulative bookings, and runs in production to inform host price recommendations. listing embeddings recursive partitioning

AI-written summary · about

" chapters (9)

1. 0:00Start
2. 2:15Overview of Airbnb
3. 4:59Pricing Airbnb
4. 7:21Market Dynamics
5. 16:20Learning Lead Time Dynamics
6. 21:37Laying the Foundation-Demand Aggregations
7. 27:25Adding Reinforcements - Structural Models
8. 29:46Final Piece to the puzzle - Forecasting the Parameters
9. 32:15Final Thoughts

" related talks

• 33mA new way to profit: Predictive analytics and dynamic pricingSrdjan Mladjenovic · Data Science Conference 4.0
• 45mKeynote: Seeking Order amidst Chaos and UncertaintyRavi Mehrotra · ODSC India 2018
• 49mBuilding Machine Learning Pipeline Using AerosolvePeng Ye · Data By the Bay 2016
• 40mMachine LearningVictor Bilyk · GDG DevFest Belarus 2018
• 22mBefore the Model: How Machine Learning Products Start...Elena Grewal · MLconf SF 2016
• 34mMachine Learning to know if you R comingClaudia Guirao · PyData Barcelona 2017
• 41mUsing Data Science in PricingVladimir Andrienko · Data Science fwdays'21 conference
• 30mReal-Time ML services: From ideation to productionIlias Sarantopoulos · PyData Eindhoven 2020