Proximity Features: Privacy-Compliant Cold-Start Personalization at Airbnb

2026-07-14 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Cybersecurity & Data Privacy · Depth: Expert, quick

Summary

Airbnb has developed Proximity Features, a privacy-compliant system designed to address cold-start personalization challenges for users with limited history, such as logged-out or first-time visitors. This system is particularly crucial for paid-channel landing pages where traditional user-level features are often unavailable. Proximity Features utilizes geo-IP data and an adaptive clustering algorithm to group approximately 1,000 geographically proximate users, generating aggregated user-level signals without requiring persistent individual identifiers during inference. Its design prioritizes privacy, operating solely on consented, aggregated data under strict privacy controls. Currently deployed in production at Airbnb for marketing landing pages and destination recommendations, with engagement email integration underway, the system has demonstrated statistically significant increases in bookings through online A/B experiments, with the most notable gains observed among users lacking sufficient historical data.

Key takeaway

For Machine Learning Engineers designing personalization systems for cold-start users or privacy-sensitive platforms, consider implementing group-based feature aggregation. Your approach could utilize geo-IP data and adaptive clustering, similar to Airbnb's Proximity Features, to generate aggregated signals for user groups. This strategy allows effective personalization without relying on persistent individual identifiers, directly addressing privacy regulations and improving engagement for users with limited history.

Key insights

Proximity Features enables privacy-compliant cold-start personalization by aggregating geo-IP data for user groups, boosting bookings for new users.

Principles

• Privacy by design using aggregated data.
• Geo-IP clustering enables cold-start personalization.

Method

The system groups users by geographic proximity using geo-IP data and an adaptive clustering algorithm, producing aggregated user-level signals for groups of ~1,000 nearby users without individual identifiers.

In practice

• Serve marketing landing pages.
• Enhance destination recommendations.
• Integrate with engagement emails.

Topics

• Cold-Start Personalization
• Privacy-Compliant ML
• Geo-IP Data
• Adaptive Clustering
• Recommendation Systems
• Airbnb

Best for: AI Engineer, AI Product Manager, CTO, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• Personalizing Airbnb search by learning from the guest journey — Transformer-based sequence models enhance search personalization by encoding comprehensive guest journey data.
• Privacy-Preserving and Verifiable Approximate Distributed Coded Computing — Model-agnostic framework unifies privacy and adversarial resilience in distributed learning via coded computing and specific defenses.
• Authentication Customer Segmentation — BEACON: K-Means Clustering — K-means clustering can segment authentication behaviors into actionable cohorts, balancing security, experience, and business outcomes.
• Airbnb Implements Context-Aware Identity Model to Support Privacy-First Social Features — Context-aware identity models enhance privacy by scoping user profiles to specific interactions.
• Data-Free Reservoir Features for Efficient Long-Horizon Cold-Start Continual Learning — CIRCLE uses fixed reservoir features and streaming linear discriminant analysis for efficient, cold-start continual learning without backbone training.
• Cracking the Cold Start Problem — Hybrid recommender systems combining collaborative filtering, embeddings, and bandit learning improve cold start and fairness.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.