In the last year Hotels.com has begun it’s journey to becoming an algorithmic business. Matt will talk about their experiences of exponential growth in Data Science Algorithms whilst at the same time the team have migrated to using Spark as their core underlying architecture from SAS / SQL, migrated to the cloud from on-premise are transforming the capability of the data science function. He will also highlight the key enablers that have made this successful including CEO support, the internal concepts of organic intelligence and how Databricks has helped make this happen. He will also highlight the pitfalls on the journey.

1. Confidential - donot distribute
Hotels.com’sjourneyto becoming
anAlgorithmicBusiness
Matthew Fryer
VP, Chief Data Science Officer
mfryer@hotels.com

2. Confidential - donot distribute
Part of Expedia, Inc. family
385,000 properties
89 countries
39 languages
>27m Hotels.com Rewards Members
Home of Captain Obvious
Billions of Recommendations, based on real-time Data per day
Hotels.com

3. Confidential - donot distribute

4. Confidential - donot distribute

5. Confidential - donot distribute
5
Data Science Engineering Front End Development

6. Confidential - donot distribute
“Artificial Intelligence Will Be
Travel’s Next Big Thing”
Barry Diller
Chairman & Senior Executive,
Expedia, Inc.
3M’s are disruptive
technology
Mobile
Messaging / NLP
Machine Learning

7. Confidential - donot distribute

8. Confidential - donot distribute
Our overall ecosystem

9. Confidential - donot distribute 9
Core Elementsof our Data ScienceCloud Platform
Databricks Unified Platform
Maestro – Our Internally Developed
Platform on AWS
(EMR, Spark, R-Studio, Intellij, SBT, Jupyter,
Zeppelin, Unit / QA, Metastore, Apache Airflow,
Keras, Tensorflow)
Proof of Concept on Google
Cloud, Beam, Spark &
Tensorflow

10. Confidential - donot distribute
DatabricksUnifiedPlatform
Chart is in1hourblocks, y axis = numberof 32coreinstances
10
• Key asset to the success of data science at Hotels.com
• Key in driving up data scientist productivity / efficiency / flexibility
• Helps make our data science lifecycle operate much easier and
faster driving speed to market
• Reliable / secure + facilitates ‘Highly Elastic’ workflows exploiting
cost effective spot instance on AWS.

11. Confidential - donot distribute
ALPs – AlgorithmLifecyclePipelineService
11

12. Confidential - donot distribute
Reference: The Influence of Visuals in Online Hotel Research and Booking Behaviour
Imagesarean importantfactorwhilechoosinga hotel
12
0% 10% 20% 30% 40% 50% 60% 70% 80%
Loyalty Program
Reviews
Hotel Brand
Star Rating
Destination Info
Images
Hotel Info
Factors other than price/location
Very Imporant/Important Important Very Important

13. Confidential - donot distribute
ComputerVisionproblemswetry to tackle
13
Near Duplicate Detection
Scene Classification Image Ranking

14. Confidential - donot distribute 14
Tagged as Bathroom

15. Confidential - donot distribute 15
GPU’s quickly became key, took a large effort to optimize using
Keras + Tensorflow (Inception v3 + ResNet)
493
67
20
7
4
1
10
100
1000
12-CPU 1-GPU 1-GPU +
limited cache
16-GPU +
limited cache
16-GPU + full
cache
Days CIFAR2
Expedia Small
15
2.5
0
10
20
16-GPU + full cache Optimized
Days

16. Confidential - donot distribute
NearDuplicateDetection:Realworldexamples
16
Non-Duplicates – probability 100%
Non-Duplicates – probability 95.91%
Duplicates – probability 97.98%
Duplicates – probability 98.43%

17. Confidential - donot distribute
ROOM/BATHROOM
Usingthe model:Real worldexamples
17
EXTERIOR/HOTEL INTERIOR/SEATING_LO
BBY
ROOM/LIVING_ROOM
ROOM/GUESTROOM
FACILITIES/DINING
INTERIOR/SEATING_LOBBY
FACILITIES/POOL

18. Confidential - donot distribute
Accuracy& ConfusionMatrix
18
• After many manual / long
winded iterations and
regularization processes
tuning hyperparameters
• We achieved good
accuracy and low
confusion matrix

19. Confidential - donot distribute
Optimizingthe photo orderfor improvedcustomer
experiences
19
Original Model
Reference: Radisson Blu Edwardian Berkshire Hotel, London

20. Confidential - donot distribute
Findingthe right hotel in our marketplace is core to
our customers needs.

21. Confidential - donot distribute
Kensington
Bloomsbury
Heathrow
Canary
Wharf
Paddington
Westminster
London City
Airport
Chelsea
Battersea
Wimbledon
Wembley
City of
London
As an exampledifferentusersegmentsliketo stayin
differentlocations

22. Confidential - donot distribute 22
Utility
Utility
Utility
just browsing! BOOK!Intent
(click)

23. Confidential - donot distribute
Thank you
mfryer@hotels.com
https://uk.linkedin.com/in/matthewfryer
@mattfryer