Personalizing Airbnb Search Through Guest Journey Learning

Airbnb has introduced a Transformer-based sequence model that encodes years of guest behavioral data to improve the user search experience. Moving away from the existing static recommendation approach, the key is to grasp the user's Journey as time-series data and surface the right accommodation at the right moment.

This model learns various behavioral sequences such as the user's past clicks, bookings, and search history to precisely capture individual preferences. Through this, it can predict not just the current search query, but the user's underlying travel style and intent.

The key features and technical approach are as follows:

• Transformer Architecture: Understands the context of user behavior by grasping long-term dependencies within sequence data
• Guest Journey Learning: Models user behavioral patterns spanning years, beyond the actions of a single session
• Personalized Search Results: Improves the accuracy of search results by combining the user's past preferences with the current context