Personalizing Airbnb Search by Learning from the Guest Journey
Using Transformer-based models to understand guest preferences and personalize search.
1Introduction
Planning a trip on Airbnb rarely happens in a single session. A guest searching for a place in San Francisco might browse dozens of listings over days, leaving a trail of views. These behaviors reveal what the guest values, but processing this data with hand-crafted features was challenging and limited.
2Challenges
Working with guest journey data presents three core challenges: data volume, sparse booking signals, and computational cost. Detecting meaningful bookings among millions of views and processing them efficiently is critical for the model's success.
3Solution: Two-Part Sequence
To tackle these challenges, we split the guest sequence into two parts: the long-term sequence includes bookings and reviews from the past seven years, while the short-term sequence captures views from the last 21 days. This approach provides both depth and recency.
4Efficient Training
To speed up model training, we used three strategies: batching searches, bucketing sequences by length, and sparse computation. These methods improved training efficiency fourfold and reduced costs.
5Serving Design
To ensure low latency in real-time searches, we processed sequences in daily batches and stored the results. When a guest searches, the model retrieves this stored data to deliver fast, personalized results.
6Advantages of Transformer Model
The Transformer-based model can represent guest preferences more richly. Unlike traditional methods, it handles complex sequences better and provides more accurate personalization.
7A/B Testing
Before launching the new model, we conducted three-week A/B tests. We monitored key business metrics and guardrails to evaluate the model's performance.
8Results
The new model helped guests find suitable listings faster. Personalized search increased booking rates and improved user experience.
9Significance
This work highlights the importance of processing large-scale behavioral data and delivering personalized experiences. For platforms like Airbnb, it's a critical step to enhance guest satisfaction and business performance.