Businesswoman discussing computer program with female colleague at desk in creative office

Machine Learning at Booking.com

Let's explore the Machine Learning world together

We are on a ground-breaking journey, working to improve the user experience for millions of travellers as they discover the world with us. 

The Booking.com Machine Learning Platform powers us to deliver personalized travel experiences for customers across the globe. Our teams use advanced ML techniques to tackle some of the most complex challenges facing travel technology, from detecting fraud to recommending the perfect property for our customers’ next stay.

Over 28 million listings
in 150,000+ destinations

Carbon neutral
offices since 2020

300+ models live in production

More than 1000 experiments taking place at any given time

Our team of experts contribute to the research academic community with regular publications
Booking.com scale data and robust ML infrastructure making it possible to get from an idea to production fast
Our decision-making is truly data-driven. We encourage our teams to think creatively, innovate and to experiment
A thriving community of hundreds of Machine Learning Scientists and Engineers from all over the world

Machine Learning Meetup recordings

Our Focus areas

Booking.com is one of the world’s online travel platforms where millions of guests find everything they need for their upcoming trip, from accommodations (including hotels, apartments, bed and breakfasts, guest houses) to flights, rental cars, and attractions. Recommendation algorithms are at the center of our product, providing customized and intelligent suggestions, and an overall better experience for the customer. In order to provide the best recommendations, we tap into a huge amount of historical information describing user engagement and preferences and develop state-of-the-art recommenders systems.

At Booking.com, we have a diverse set of textual data sources spanning over 42 languages, different business verticals (e.g. accommodations, food, flights, cars), multiple sub-domains (e.g. user reviews, partner provided descriptions, messages) and also varying levels of noise. Our Natural Language Processing (NLP) group is building methodologies to best leverage this content efficiently which is subsequently powering many user applications like ranking, search, moderation, intelligent assistance, translations etc.

Images can help with everything from selecting the perfect location to selecting the right hotel, and even deciding on the best attraction. Our travel products come with a wealth of visual content, as you can imagine and machine learning helps us surface relevant content to customers in the right context.

Multi-Armed Bandits (MAB) have gained increased popularity during recent years in e-commerce companies. At Booking.com we developed online implementations of contextual Multi Armed Bandit (MAB) algorithms to optimize the layout of our pages and for our email marketing campaigns to enable us to send personalized and relevant messages to our customers to empower them to travel.

MAB algorithm and reinforcement learning algorithms are also used for our email marketing campaigns, enabling us to send personalised and relevant messages to our customers to empower them to travel.

Deploying ML use cases in production at Booking.com scale is a challenging task: Our ML engineers work together with the ML scientists to build and release business critical ML models in areas such as ranking, fraud prevention, performance marketing, and personalisation of users’ journey. On top of that, other ML engineers have built over the years a Machine Learning Platform with the ambition to support the life cycle of every internal ML project.

Our people's perspective

Publications and articles

Young Caucasian man riding on boat through Mekong delta in Vietnam

A/B/n testing with control in the presence of subpopulations

18254_org

Machine Learning in production: the Booking.com approach

Group of friends travelling with a vintage mini van

Personalization in Practice