Speech deliverd on 20 June 2020 at TR.AI Meetup, Istanbul TR.AI Türkiye Yapay Zeka İnisiyatifi AI/ML PoweredPersonalized Recommendations in Gaming Industry Amazon Web Services - AWS

1. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
AI/ML Powered
Personalized Recommendations
in Gaming Industry
20 June 2020, TR AI Meetup, Istanbul (Virtual)
Hasan Basri AKIRMAK, MSc., Exec-MBA
Amazon Web Services
TR AI Türkiye Yapay Zeka İnisiyatifi

2. © 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Agenda
• AWS in Game Tech
• Evolution of Algorithms
• Challenges of Personalized Recommendations
• Customer Reference: Voodoo

3. © 2020, Amazon Web Services, Inc. or its Affiliates.
About Amazon: Game Tech Customers

4. © 2020, Amazon Web Services, Inc. or its Affiliates.
About Amazon: Game Tech Customers
For more case studies:
refer to: https://aws.amazon.com/tr/gametech/

5. © 2020, Amazon Web Services, Inc. or its Affiliates.
Using AI to Personalize Video
Games Experience

6. © 2020, Amazon Web Services, Inc. or its Affiliates.
Effective personalization involves solving multiple hard
problems
Popularity trap
Naïve models give recommendations similar to popular items
Cold starts
New users should get relevant recommendations, new items should show in recommendations
Scale
Recommendations should scale across millions of users and items
Real time
Personalization must be responsive to the changing user intent
Custom models
Personalization models must accurately reflect business context and user behavior

7. © 2020, Amazon Web Services, Inc. or its Affiliates.
History of Personalization at Amazon
Item Based Collaborative Filtering (2003)
«By the time we published in IEEE in 2003, item-based collaborative
filtering was widely deployed across Amazon.com.
Others have reported using the algorithm, too.
In 2010, YouTube reported using it for recommending videos.
Many open source and third-party vendors included the algorithm, and it
showed up widely in online retail, travel, news, advertising, and more.
The recommendations were used so extensively by Amazon.com that a
report estimated 30% of Amazon.com’s page views were from
recommendations.
Similarly, Netflix used recommender systems so extensively that their
Chief Product Officer, Neil Hunt, indicated that more than 80% of movies
watched on Netflix came through recommendations.»
[Greg Linden, et.al.]

8. © 2020, Amazon Web Services, Inc. or its Affiliates.
Personalization at Amazon
User personalization
• When to use:
• Landing pages
• Cart recommendations
• Email promotions (often preprocessed)
• Detail pages in addition to similar
items
Similar items
• Up- sell recommendations (SIMS +
business rules)
Personalized ranking
• Category-based recommendations
• Recommendations under business
constraints (e.g., recommend only from
free-to-watch movies)

9. © 2020, Amazon Web Services, Inc. or its Affiliates.
ML & Personalization

10. © 2020, Amazon Web Services, Inc. or its Affiliates.
Evolution of Algorithms
Rolling
Average
T-SVD [2009] PMF [2008] RRN [2017] DeepRec
[2017]
HRNN
Ratings RMSE on Netflix
98 MM interactions, 500k users, 18k items
Rolling Average T-SVD [2009] PMF [2008]
RRN [2017] DeepRec [2017] HRNN
Rolling Average FM [2012] I-AutoRec
[2015]
RNN HRNN
Ratings RMSE on MovieLens
20 MM interactions, 173k users, 131k items
Rolling Average FM [2012]
I-AutoRec [2015] RNN
HRNN is one of the more advanced algorithms provided by Amazon Personalize. Beyond better performance, it supports personalization of
the items for a specific user based on their past behavior and can intake real time events in order to alter recommendations for a user
without retraining.

11. © 2020, Amazon Web Services, Inc. or its Affiliates.
Item Recommendation with Factorization Machines (2012)
• A factorization machine is a general-
purpose supervised learning
algorithm that you can use for both
classification and regression tasks.
• It is designed to capture interactions
between features within high
dimensional sparse datasets
economically.
• Factorization machines are a good
choice for tasks, such as click
prediction and item
recommendation.
FM on SageMaker: https://docs.aws.amazon.com/sagemaker/latest/dg/fact-machines.html
Image Credit: Factorization Machines Steffen Rendle, et. Al.
https://www.csie.ntu.edu.tw/~b97053/paper/Rendle2010FM.pdf

12. © 2020, Amazon Web Services, Inc. or its Affiliates.
Extending Factorization Machines to predict top N Items
• When you apply a recommendation model, you
often want to provide a user as input and receive
a list of the top x items that best match the user’s
preferences.
• When the number of items is moderate, you can
do this by querying the model for user, item for
all possible items. However, this approach doesn’t
scale well when the number of items is large.
• In this scenario, you can use the
Amazon SageMaker k-nearest neighbors (k-NN)
algorithm to speed up top x prediction tasks.
.
Factorizati
on
machines
model
Repacka
ge model
data
k-NN
model
Top X
results
FM with K-NN for Bacth. Sample Notebook: https://github.com/aws-samples/amazon-sagemaker-
architecting-for-ml/blob/master/Starter-Code/Recommendation-System-FM-KNN.ipynb

13. © 2020, Amazon Web Services, Inc. or its Affiliates.
User personalization with Sequence Models
Recurrent Neural Networks (RNN) Models
ü Assigns more attention to recent events
ü Real-time interactive responses
Other
information
Learned user
representation
(Order and timing matter)
(recurrent neural network)

14. © 2020, Amazon Web Services, Inc. or its Affiliates.
Customer Reference

15. © 2020, Amazon Web Services, Inc. or its Affiliates.
Voodoo ML-based personalized recommendations
• The number one gaming company
worldwide with the most downloads on the
App Store with 3.7 billion downloads.
• Decide in real time which ad to show to their
players
• Close to 1 Billion predictions / day.
• With AWS machine learning, Voodoo put an
accurate model into production in less than
a week.
• Using Amazon SageMaker &Amazon
Personalize for ML

16. © 2020, Amazon Web Services, Inc. or its Affiliates.
Voodoo ML-based personalized recommendations
AWS re:Invent 2019: Delight your customers with ML-based personalized
recommendations (AIM323)
Wath the session delivered by Voodoo CTO at
https://www.youtube.com/watch?v=XCE3PnGb3As

17. © 2020, Amazon Web Services, Inc. or its Affiliates.
The AWS ML Stack
Broadest and most complete set of Machine Learning capabilities
VISION SPEECH TEXT SEARCH NEW CHATBOTS PERSONALIZATION FORECASTING FRAUD NEW DEVELOPMENT NEW CONTACT CENTERS NEW
Amazon SageMaker Ground
Truth
Augmented
AI
SageMaker
Neo
Built-in
algorithms
SageMaker
Notebooks NEW
SageMaker
Experiments NEW
Model
tuning
SageMaker
Debugger NEW
SageMaker
Autopilot NEW
Model
hosting
SageMaker
Model Monitor NEW
Deep Learning
AMIs & Containers
GPUs &
CPUs
Elastic
Inference
Inferentia FPGA
Amazon
Rekognition
Amazon
Polly
Amazon
Transcribe
+Medical
Amazon
Comprehend
+Medical
Amazon
Translate
Amazon
Lex
Amazon
Personalize
Amazon
Forecast
Amazon
Fraud Detector
Amazon
CodeGuru
AI SERVICES
ML SERVICES
ML FRAMEWORKS & INFRASTRUCTURE
Amazon
Textract
Amazon
Kendra
Contact Lens
For Amazon Connect
SageMaker Studio IDE NEW
NEW
NEW
NEW
NEW

18. © 2020, Amazon Web Services, Inc. or its Affiliates.
© 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Factorization Machines
Log_loss F1 Score Seconds
Amazon
SageMaker
0.494 0.277 820
Other (10 Iter) 0.516 0.190 650
Other (20 Iter) 0.507 0.254 1300
Other (50 Iter) 0.481 0.313 3250
˜y = w0 + hw1, xi +
X
i,j>i
xixj · hvi, vji
Test Scenario: Click Prediction 1 TB advertising dataset, m4.4xlarge machines, perfect scaling.
$-
$20,00
$40,00
$60,00
$80,00
$100,00
$120,00
$140,00
$160,00
$180,00
$200,00
1 2 3 4 5 6 7 8
CostinDollars Billable Time in Hours
10
machines
20
machines
30
machines
4050
SageMaker Built-in Algoritm Details (Dev Guide): https://docs.aws.amazon.com/sagemaker/latest/dg/fact-machines.html

19. © 2020, Amazon Web Services, Inc. or its Affiliates.
Real-time Works with almost any
product or content
K E Y F E A T U R E S
Responsive to changes
in intent
Automated
machine learning
Deliver high-quality
recommendations
Deep learning
algorithms
Very fast to setup
Delight your customers
Improve customer experience with Amazon Personalize

20. © 2020, Amazon Web Services, Inc. or its Affiliates.
Customized
personalization
API
Amazon Personalize
Inspect
data
Identify
features
Select
hyper-
parameters
Train
models
Optimize
models
Host
models
Real-time
feature
store
Amazon Personalize
Behind the scenes
Fully managed by Amazon Personalize
Item metadata
(details of articles,
products, videos, etc).
User metadata
(age, location, etc.)
User events/
interactions
(views, signups,
conversions, etc.)

21. © 2020, Amazon Web Services, Inc. or its Affiliates.
Turkey Time:
19:00-23:00
today
https://aws.amazon.com/gametech/events/digital-download-online/

22. © 2020, Amazon Web Services, Inc. or its Affiliates.
Q&A
AI Powered Recommendation Systems