Machine Learning-Powered Search Ranking of Airbnb Experiences
Source: https://medium.com/airbnb-engineering/machine-learning-powered-search-ranking-of-airbnb-experiences-110b4b1a0789 Authors: Mihajlo Grbovic, Eric Wu, Pai Liu, Chun How Tan, Liang Wu, Bo Yu, Alex Tian
Summary
Case study of how Airbnb iterated through four stages of ML-powered search ranking for Experiences, achieving ~28% cumulative booking improvement.
Four-Stage Progression
Stage 1 — Baseline GBDT (+13% bookings)
Gradient Boosted Decision Trees with binary classification on 50,000 labeled examples. Features: duration, price, reviews, booking counts.
Stage 2 — Personalization (+7.9%)
Added two dimensions:
- Features from booked homes (distance, trip date availability)
- User click history (category intensity/recency, time-of-day preferences)
Offline pre-computation of rankings for 1M+ active users daily.
Stage 3 — Online Scoring (+5.1%)
Real-time inference with query features: distance to entered location, guest count, browser language, origin-destination preferences. 2M+ training examples, 90 features.
Stage 4 — Business Rules (+2.2%)
Weighted training data to promote: quality (5-star rebooking 1.5x more), emerging hits (+14%), category diversity (+2.3%).
Key Features
- Experience features: reviews, booking velocity, occupancy, duration, price, category
- Query features: distance, guest count, language availability, geographic preferences
- Personalization: Category Intensity = weighted sum of user clicks with recency decay
Evaluation
Offline: AUC and NDCG; Online: A/B testing for booking rate.