SlideShare a Scribd company logo
1 of 31
Download to read offline
Building a
Personalized Messaging
System at Netflix
Grace Huang
July 31th, 2020
Data Council SF 2020
When members find content
they love, they enjoy our
service more
Our messages are designed to
help them find content to enjoy
Some ways to reach out to our members
Email Push SMS
In-App: Notifications and Alerts
A Variety of Message Types
Recommendations New Arrival New Season Alert Coming Soon
Candidate Messages
Heuristics based:
● If a user watched Stranger Things
Season 1, then send a message
about Season 2 arrival
● Do not send if a user has had a
similar recommendation in the past
x days
● ….
Decision Engine
Email
Push Notifications
In-App Alerts
A Heuristics Driven Paradigm
Candidate Messages
ML-Driven
Decision Engine
Email
Push Notifications
In-App Alerts
A Heuristics Driven Paradigm
How many to send?
What to send?
Key Considerations for the System
● Making a personalized, timely decision for every Netflix subscriber
● Removing bias from the system
● Maximizing causal impact
● Balancing reward against cost
1
Personalizing the
messaging decision
NETFLIX
p(Y|X)
A Personalized Messaging Decision
Can be estimated using a variety of classification (or regression) techniques -
Linear (or Logistic)Regression, GBDT, Neural Network...etc
Outcome Features
But...
How to obtain data with the full range of messaging frequency
and message type variations?
2
Removing Bias
from the System
NETFLIX
The Obvious Candidate: Explore/Exploit
ε-greedy UCB Thompson
Sampling
chart source
● Take a random sample from
each arm's PDF
● Choose the arm with the
highest sampled value.
● Explore with probability ε
● Otherwise, choose arm
with best action
● Pull arm with the highest
upper confidence bound
An Example Approach: Personalized
Messaging using Contextual Bandit
Context Messages Chosen
Choose a random message:
Other Examples of Debiasing Techniques:
Propensity Correction
● Instrument Variables..
R YZ
Noise OutcomeRec
● Propensity Correction (e.g. IPS)
But...
Did a subscriber visit Netflix and watch a movie because the
message we sent was truly relevant and helpful?
Would they have watched a show even if we did not reach out?
3
Maximizing Causal
Impact
NETFLIX
● p(Y|X) only captures
correlation
● Every observation is
influenced by past actions
from our messaging system
A Causal Personalization System
https://xkcd.com/552/
p(Y|X)
Recall that we built a Correlational
Personalization Model for Messaging...
Outcome Features
p(Y|X, do(R))
Explicitly Model Past Actions
R∈ {∅, [ ], [ ], ….}
Personalized Decision Making
>
Member satisfaction with message Member satisfaction with no message
Send a Message when...
p(Y|X, do(∅))p(Y|X, do([ ] ))
4
Balancing Reward
against Cost
NETFLIX
More ≠ Better
How to impose a volume constraint?
How to Impose a Volume Constraint?
● A simple approach:
Set an Incrementality Threshold
● Other flavors of
Reinforcement Learning
An example of how to put this together….
A Causal Bandit
● Randomly sample from
frequency and candidate
message distributions
● Estimate incrementality
of a message
● Greedily assemble the
set of messages to
send, subject to an
incrementality threshold
Offline Evaluation
Simulate online metrics offline!
User 1 User 2 User 3 User 4 User 5 User 6
Explore policy
Reward
Policy to evaluate
● An A/B test still allows us to evaluate the long term behavior of a given policy
Evaluating Online
Non-personalized
Personalized variant 1
Personalized variant 2
Personalized variant 3
Recap
● Making a personalized, timely decision for every Netflix subscriber
● Removing bias from the system
● Maximizing causal impact
● Balancing reward against cost
● Lots of exciting future work…..
Check out more in Recap: Designing a more Efficient Estimator for
Off-policy Evaluation in Bandits with Large Action Spaces
Grace Huang
ghuang@netflix.com
Thank
You.

More Related Content

What's hot

Artwork Personalization at Netflix
Artwork Personalization at NetflixArtwork Personalization at Netflix
Artwork Personalization at NetflixJustin Basilico
 
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
 Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se... Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...Sudeep Das, Ph.D.
 
Missing values in recommender models
Missing values in recommender modelsMissing values in recommender models
Missing values in recommender modelsParmeshwar Khurd
 
Déjà Vu: The Importance of Time and Causality in Recommender Systems
Déjà Vu: The Importance of Time and Causality in Recommender SystemsDéjà Vu: The Importance of Time and Causality in Recommender Systems
Déjà Vu: The Importance of Time and Causality in Recommender SystemsJustin Basilico
 
Personalizing "The Netflix Experience" with Deep Learning
Personalizing "The Netflix Experience" with Deep LearningPersonalizing "The Netflix Experience" with Deep Learning
Personalizing "The Netflix Experience" with Deep LearningAnoop Deoras
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsYves Raimond
 
Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations at NetflixLinas Baltrunas
 
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...MLconf
 
Crafting Recommenders: the Shallow and the Deep of it!
Crafting Recommenders: the Shallow and the Deep of it! Crafting Recommenders: the Shallow and the Deep of it!
Crafting Recommenders: the Shallow and the Deep of it! Sudeep Das, Ph.D.
 
Tutorial on Deep Learning in Recommender System, Lars summer school 2019
Tutorial on Deep Learning in Recommender System, Lars summer school 2019Tutorial on Deep Learning in Recommender System, Lars summer school 2019
Tutorial on Deep Learning in Recommender System, Lars summer school 2019Anoop Deoras
 
Calibrated Recommendations
Calibrated RecommendationsCalibrated Recommendations
Calibrated RecommendationsHarald Steck
 
Recommender Systems In Industry
Recommender Systems In IndustryRecommender Systems In Industry
Recommender Systems In IndustryXavier Amatriain
 
Boston ML - Architecting Recommender Systems
Boston ML - Architecting Recommender SystemsBoston ML - Architecting Recommender Systems
Boston ML - Architecting Recommender SystemsJames Kirk
 
Shallow and Deep Latent Models for Recommender System
Shallow and Deep Latent Models for Recommender SystemShallow and Deep Latent Models for Recommender System
Shallow and Deep Latent Models for Recommender SystemAnoop Deoras
 
Personalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsPersonalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsJustin Basilico
 
Netflix talk at ML Platform meetup Sep 2019
Netflix talk at ML Platform meetup Sep 2019Netflix talk at ML Platform meetup Sep 2019
Netflix talk at ML Platform meetup Sep 2019Faisal Siddiqi
 
Recommendation at Netflix Scale
Recommendation at Netflix ScaleRecommendation at Netflix Scale
Recommendation at Netflix ScaleJustin Basilico
 
RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020
RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020
RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020Zachary Schendel
 
Making Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableMaking Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableJustin Basilico
 

What's hot (20)

Artwork Personalization at Netflix
Artwork Personalization at NetflixArtwork Personalization at Netflix
Artwork Personalization at Netflix
 
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
 Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se... Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
 
Missing values in recommender models
Missing values in recommender modelsMissing values in recommender models
Missing values in recommender models
 
Déjà Vu: The Importance of Time and Causality in Recommender Systems
Déjà Vu: The Importance of Time and Causality in Recommender SystemsDéjà Vu: The Importance of Time and Causality in Recommender Systems
Déjà Vu: The Importance of Time and Causality in Recommender Systems
 
Personalizing "The Netflix Experience" with Deep Learning
Personalizing "The Netflix Experience" with Deep LearningPersonalizing "The Netflix Experience" with Deep Learning
Personalizing "The Netflix Experience" with Deep Learning
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations at Netflix
 
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
 
Crafting Recommenders: the Shallow and the Deep of it!
Crafting Recommenders: the Shallow and the Deep of it! Crafting Recommenders: the Shallow and the Deep of it!
Crafting Recommenders: the Shallow and the Deep of it!
 
Tutorial on Deep Learning in Recommender System, Lars summer school 2019
Tutorial on Deep Learning in Recommender System, Lars summer school 2019Tutorial on Deep Learning in Recommender System, Lars summer school 2019
Tutorial on Deep Learning in Recommender System, Lars summer school 2019
 
Calibrated Recommendations
Calibrated RecommendationsCalibrated Recommendations
Calibrated Recommendations
 
Recommender Systems In Industry
Recommender Systems In IndustryRecommender Systems In Industry
Recommender Systems In Industry
 
Learning to Personalize
Learning to PersonalizeLearning to Personalize
Learning to Personalize
 
Boston ML - Architecting Recommender Systems
Boston ML - Architecting Recommender SystemsBoston ML - Architecting Recommender Systems
Boston ML - Architecting Recommender Systems
 
Shallow and Deep Latent Models for Recommender System
Shallow and Deep Latent Models for Recommender SystemShallow and Deep Latent Models for Recommender System
Shallow and Deep Latent Models for Recommender System
 
Personalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsPersonalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing Recommendations
 
Netflix talk at ML Platform meetup Sep 2019
Netflix talk at ML Platform meetup Sep 2019Netflix talk at ML Platform meetup Sep 2019
Netflix talk at ML Platform meetup Sep 2019
 
Recommendation at Netflix Scale
Recommendation at Netflix ScaleRecommendation at Netflix Scale
Recommendation at Netflix Scale
 
RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020
RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020
RecSys 2020 A Human Perspective on Algorithmic Similarity Schendel 9-2020
 
Making Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableMaking Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms Reliable
 

Similar to Data council SF 2020 Building a Personalized Messaging System at Netflix

Artificial Intelligence at LinkedIn
Artificial Intelligence at LinkedInArtificial Intelligence at LinkedIn
Artificial Intelligence at LinkedInBill Liu
 
Fairness, Transparency, and Privacy in AI @LinkedIn
Fairness, Transparency, and Privacy in AI @LinkedInFairness, Transparency, and Privacy in AI @LinkedIn
Fairness, Transparency, and Privacy in AI @LinkedInC4Media
 
Creating a customer segmentation workflow with knime
Creating a customer segmentation workflow with knimeCreating a customer segmentation workflow with knime
Creating a customer segmentation workflow with knimeKnoldus Inc.
 
Machine Learning in e commerce - Reboot
Machine Learning in e commerce - RebootMachine Learning in e commerce - Reboot
Machine Learning in e commerce - RebootMarion DE SOUSA
 
QuestBack - International Presentation
QuestBack - International PresentationQuestBack - International Presentation
QuestBack - International PresentationFalco Noort
 
Machine Learning AND Deep Learning for OpenPOWER
Machine Learning AND Deep Learning for OpenPOWERMachine Learning AND Deep Learning for OpenPOWER
Machine Learning AND Deep Learning for OpenPOWERGanesan Narayanasamy
 
Introducing Amazon Pinpoint – Targeted Push Notifications for Mobile Apps
Introducing Amazon Pinpoint – Targeted Push Notifications for Mobile AppsIntroducing Amazon Pinpoint – Targeted Push Notifications for Mobile Apps
Introducing Amazon Pinpoint – Targeted Push Notifications for Mobile AppsAmazon Web Services
 
DataScientist Job : Between Myths and Reality.pdf
DataScientist Job : Between Myths and Reality.pdfDataScientist Job : Between Myths and Reality.pdf
DataScientist Job : Between Myths and Reality.pdfJedha Bootcamp
 
Fairness, Transparency, and Privacy in AI @ LinkedIn
Fairness, Transparency, and Privacy in AI @ LinkedInFairness, Transparency, and Privacy in AI @ LinkedIn
Fairness, Transparency, and Privacy in AI @ LinkedInKrishnaram Kenthapadi
 
A Scalable, High-performance Algorithm for Hybrid Job Recommendations
A Scalable, High-performance Algorithm for Hybrid Job RecommendationsA Scalable, High-performance Algorithm for Hybrid Job Recommendations
A Scalable, High-performance Algorithm for Hybrid Job RecommendationsToon De Pessemier
 
IRIS.TV Talks Future of Video Personalization at Cross Campus LA
IRIS.TV Talks Future of Video Personalization at Cross Campus LAIRIS.TV Talks Future of Video Personalization at Cross Campus LA
IRIS.TV Talks Future of Video Personalization at Cross Campus LAIRIS.TV
 
Understanding and maintaining your market to maximise revenue generation opp...
Understanding and maintaining your market to maximise revenue generation opp...Understanding and maintaining your market to maximise revenue generation opp...
Understanding and maintaining your market to maximise revenue generation opp...Miguel Simões
 
Use of Analytics by Netflix - Case Study
Use of Analytics by Netflix - Case StudyUse of Analytics by Netflix - Case Study
Use of Analytics by Netflix - Case StudySaket Toshniwal
 
使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群
使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群
使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群Amazon Web Services
 
Deepak-Computational Advertising-The LinkedIn Way
Deepak-Computational Advertising-The LinkedIn WayDeepak-Computational Advertising-The LinkedIn Way
Deepak-Computational Advertising-The LinkedIn Wayyingfeng
 
Product marketing AV concept options
Product marketing AV concept optionsProduct marketing AV concept options
Product marketing AV concept optionsMomy Saikia
 
MailNinja Managed Services 2015
MailNinja Managed Services 2015MailNinja Managed Services 2015
MailNinja Managed Services 2015MailNinja
 

Similar to Data council SF 2020 Building a Personalized Messaging System at Netflix (20)

Artificial Intelligence at LinkedIn
Artificial Intelligence at LinkedInArtificial Intelligence at LinkedIn
Artificial Intelligence at LinkedIn
 
Fairness, Transparency, and Privacy in AI @LinkedIn
Fairness, Transparency, and Privacy in AI @LinkedInFairness, Transparency, and Privacy in AI @LinkedIn
Fairness, Transparency, and Privacy in AI @LinkedIn
 
Creating a customer segmentation workflow with knime
Creating a customer segmentation workflow with knimeCreating a customer segmentation workflow with knime
Creating a customer segmentation workflow with knime
 
Quest Back 2010
Quest Back 2010Quest Back 2010
Quest Back 2010
 
Machine Learning in e commerce - Reboot
Machine Learning in e commerce - RebootMachine Learning in e commerce - Reboot
Machine Learning in e commerce - Reboot
 
QuestBack - International Presentation
QuestBack - International PresentationQuestBack - International Presentation
QuestBack - International Presentation
 
Quest Back 2010 En
Quest Back 2010 EnQuest Back 2010 En
Quest Back 2010 En
 
Machine Learning AND Deep Learning for OpenPOWER
Machine Learning AND Deep Learning for OpenPOWERMachine Learning AND Deep Learning for OpenPOWER
Machine Learning AND Deep Learning for OpenPOWER
 
Introducing Amazon Pinpoint – Targeted Push Notifications for Mobile Apps
Introducing Amazon Pinpoint – Targeted Push Notifications for Mobile AppsIntroducing Amazon Pinpoint – Targeted Push Notifications for Mobile Apps
Introducing Amazon Pinpoint – Targeted Push Notifications for Mobile Apps
 
DataScientist Job : Between Myths and Reality.pdf
DataScientist Job : Between Myths and Reality.pdfDataScientist Job : Between Myths and Reality.pdf
DataScientist Job : Between Myths and Reality.pdf
 
Fairness, Transparency, and Privacy in AI @ LinkedIn
Fairness, Transparency, and Privacy in AI @ LinkedInFairness, Transparency, and Privacy in AI @ LinkedIn
Fairness, Transparency, and Privacy in AI @ LinkedIn
 
A Scalable, High-performance Algorithm for Hybrid Job Recommendations
A Scalable, High-performance Algorithm for Hybrid Job RecommendationsA Scalable, High-performance Algorithm for Hybrid Job Recommendations
A Scalable, High-performance Algorithm for Hybrid Job Recommendations
 
IRIS.TV Talks Future of Video Personalization at Cross Campus LA
IRIS.TV Talks Future of Video Personalization at Cross Campus LAIRIS.TV Talks Future of Video Personalization at Cross Campus LA
IRIS.TV Talks Future of Video Personalization at Cross Campus LA
 
Understanding and maintaining your market to maximise revenue generation opp...
Understanding and maintaining your market to maximise revenue generation opp...Understanding and maintaining your market to maximise revenue generation opp...
Understanding and maintaining your market to maximise revenue generation opp...
 
Use of Analytics by Netflix - Case Study
Use of Analytics by Netflix - Case StudyUse of Analytics by Netflix - Case Study
Use of Analytics by Netflix - Case Study
 
使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群
使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群
使用 Amazon Pinpoint 讓你的行動 App 更精準接觸客群
 
Deepak-Computational Advertising-The LinkedIn Way
Deepak-Computational Advertising-The LinkedIn WayDeepak-Computational Advertising-The LinkedIn Way
Deepak-Computational Advertising-The LinkedIn Way
 
GT_feed
GT_feedGT_feed
GT_feed
 
Product marketing AV concept options
Product marketing AV concept optionsProduct marketing AV concept options
Product marketing AV concept options
 
MailNinja Managed Services 2015
MailNinja Managed Services 2015MailNinja Managed Services 2015
MailNinja Managed Services 2015
 

Recently uploaded

Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365
Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365
Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 3652toLead Limited
 
How to Remove Document Management Hurdles with X-Docs?
How to Remove Document Management Hurdles with X-Docs?How to Remove Document Management Hurdles with X-Docs?
How to Remove Document Management Hurdles with X-Docs?XfilesPro
 
Presentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreterPresentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreternaman860154
 
Slack Application Development 101 Slides
Slack Application Development 101 SlidesSlack Application Development 101 Slides
Slack Application Development 101 Slidespraypatel2
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerThousandEyes
 
Integration and Automation in Practice: CI/CD in Mule Integration and Automat...
Integration and Automation in Practice: CI/CD in Mule Integration and Automat...Integration and Automation in Practice: CI/CD in Mule Integration and Automat...
Integration and Automation in Practice: CI/CD in Mule Integration and Automat...Patryk Bandurski
 
08448380779 Call Girls In Diplomatic Enclave Women Seeking Men
08448380779 Call Girls In Diplomatic Enclave Women Seeking Men08448380779 Call Girls In Diplomatic Enclave Women Seeking Men
08448380779 Call Girls In Diplomatic Enclave Women Seeking MenDelhi Call girls
 
Install Stable Diffusion in windows machine
Install Stable Diffusion in windows machineInstall Stable Diffusion in windows machine
Install Stable Diffusion in windows machinePadma Pradeep
 
Salesforce Community Group Quito, Salesforce 101
Salesforce Community Group Quito, Salesforce 101Salesforce Community Group Quito, Salesforce 101
Salesforce Community Group Quito, Salesforce 101Paola De la Torre
 
Human Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR SystemsHuman Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR SystemsMark Billinghurst
 
08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking MenDelhi Call girls
 
Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024BookNet Canada
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationSafe Software
 
Breaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path MountBreaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path MountPuma Security, LLC
 
#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024BookNet Canada
 
Maximizing Board Effectiveness 2024 Webinar.pptx
Maximizing Board Effectiveness 2024 Webinar.pptxMaximizing Board Effectiveness 2024 Webinar.pptx
Maximizing Board Effectiveness 2024 Webinar.pptxOnBoard
 
AI as an Interface for Commercial Buildings
AI as an Interface for Commercial BuildingsAI as an Interface for Commercial Buildings
AI as an Interface for Commercial BuildingsMemoori
 
Understanding the Laravel MVC Architecture
Understanding the Laravel MVC ArchitectureUnderstanding the Laravel MVC Architecture
Understanding the Laravel MVC ArchitecturePixlogix Infotech
 
A Domino Admins Adventures (Engage 2024)
A Domino Admins Adventures (Engage 2024)A Domino Admins Adventures (Engage 2024)
A Domino Admins Adventures (Engage 2024)Gabriella Davis
 
My Hashitalk Indonesia April 2024 Presentation
My Hashitalk Indonesia April 2024 PresentationMy Hashitalk Indonesia April 2024 Presentation
My Hashitalk Indonesia April 2024 PresentationRidwan Fadjar
 

Recently uploaded (20)

Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365
Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365
Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365
 
How to Remove Document Management Hurdles with X-Docs?
How to Remove Document Management Hurdles with X-Docs?How to Remove Document Management Hurdles with X-Docs?
How to Remove Document Management Hurdles with X-Docs?
 
Presentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreterPresentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreter
 
Slack Application Development 101 Slides
Slack Application Development 101 SlidesSlack Application Development 101 Slides
Slack Application Development 101 Slides
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected Worker
 
Integration and Automation in Practice: CI/CD in Mule Integration and Automat...
Integration and Automation in Practice: CI/CD in Mule Integration and Automat...Integration and Automation in Practice: CI/CD in Mule Integration and Automat...
Integration and Automation in Practice: CI/CD in Mule Integration and Automat...
 
08448380779 Call Girls In Diplomatic Enclave Women Seeking Men
08448380779 Call Girls In Diplomatic Enclave Women Seeking Men08448380779 Call Girls In Diplomatic Enclave Women Seeking Men
08448380779 Call Girls In Diplomatic Enclave Women Seeking Men
 
Install Stable Diffusion in windows machine
Install Stable Diffusion in windows machineInstall Stable Diffusion in windows machine
Install Stable Diffusion in windows machine
 
Salesforce Community Group Quito, Salesforce 101
Salesforce Community Group Quito, Salesforce 101Salesforce Community Group Quito, Salesforce 101
Salesforce Community Group Quito, Salesforce 101
 
Human Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR SystemsHuman Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR Systems
 
08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men
 
Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
 
Breaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path MountBreaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path Mount
 
#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
#StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
 
Maximizing Board Effectiveness 2024 Webinar.pptx
Maximizing Board Effectiveness 2024 Webinar.pptxMaximizing Board Effectiveness 2024 Webinar.pptx
Maximizing Board Effectiveness 2024 Webinar.pptx
 
AI as an Interface for Commercial Buildings
AI as an Interface for Commercial BuildingsAI as an Interface for Commercial Buildings
AI as an Interface for Commercial Buildings
 
Understanding the Laravel MVC Architecture
Understanding the Laravel MVC ArchitectureUnderstanding the Laravel MVC Architecture
Understanding the Laravel MVC Architecture
 
A Domino Admins Adventures (Engage 2024)
A Domino Admins Adventures (Engage 2024)A Domino Admins Adventures (Engage 2024)
A Domino Admins Adventures (Engage 2024)
 
My Hashitalk Indonesia April 2024 Presentation
My Hashitalk Indonesia April 2024 PresentationMy Hashitalk Indonesia April 2024 Presentation
My Hashitalk Indonesia April 2024 Presentation
 

Data council SF 2020 Building a Personalized Messaging System at Netflix

  • 1. Building a Personalized Messaging System at Netflix Grace Huang July 31th, 2020 Data Council SF 2020
  • 2.
  • 3. When members find content they love, they enjoy our service more
  • 4. Our messages are designed to help them find content to enjoy
  • 5. Some ways to reach out to our members Email Push SMS
  • 7. A Variety of Message Types Recommendations New Arrival New Season Alert Coming Soon
  • 8. Candidate Messages Heuristics based: ● If a user watched Stranger Things Season 1, then send a message about Season 2 arrival ● Do not send if a user has had a similar recommendation in the past x days ● …. Decision Engine Email Push Notifications In-App Alerts A Heuristics Driven Paradigm
  • 9. Candidate Messages ML-Driven Decision Engine Email Push Notifications In-App Alerts A Heuristics Driven Paradigm How many to send? What to send?
  • 10. Key Considerations for the System ● Making a personalized, timely decision for every Netflix subscriber ● Removing bias from the system ● Maximizing causal impact ● Balancing reward against cost
  • 12. p(Y|X) A Personalized Messaging Decision Can be estimated using a variety of classification (or regression) techniques - Linear (or Logistic)Regression, GBDT, Neural Network...etc Outcome Features
  • 13. But... How to obtain data with the full range of messaging frequency and message type variations?
  • 14. 2 Removing Bias from the System NETFLIX
  • 15. The Obvious Candidate: Explore/Exploit ε-greedy UCB Thompson Sampling chart source ● Take a random sample from each arm's PDF ● Choose the arm with the highest sampled value. ● Explore with probability ε ● Otherwise, choose arm with best action ● Pull arm with the highest upper confidence bound
  • 16. An Example Approach: Personalized Messaging using Contextual Bandit Context Messages Chosen Choose a random message:
  • 17. Other Examples of Debiasing Techniques: Propensity Correction ● Instrument Variables.. R YZ Noise OutcomeRec ● Propensity Correction (e.g. IPS)
  • 18. But... Did a subscriber visit Netflix and watch a movie because the message we sent was truly relevant and helpful? Would they have watched a show even if we did not reach out?
  • 20. ● p(Y|X) only captures correlation ● Every observation is influenced by past actions from our messaging system A Causal Personalization System https://xkcd.com/552/
  • 21. p(Y|X) Recall that we built a Correlational Personalization Model for Messaging... Outcome Features
  • 22. p(Y|X, do(R)) Explicitly Model Past Actions R∈ {∅, [ ], [ ], ….} Personalized Decision Making
  • 23. > Member satisfaction with message Member satisfaction with no message Send a Message when... p(Y|X, do(∅))p(Y|X, do([ ] ))
  • 25. More ≠ Better How to impose a volume constraint?
  • 26. How to Impose a Volume Constraint? ● A simple approach: Set an Incrementality Threshold ● Other flavors of Reinforcement Learning
  • 27. An example of how to put this together…. A Causal Bandit ● Randomly sample from frequency and candidate message distributions ● Estimate incrementality of a message ● Greedily assemble the set of messages to send, subject to an incrementality threshold
  • 28. Offline Evaluation Simulate online metrics offline! User 1 User 2 User 3 User 4 User 5 User 6 Explore policy Reward Policy to evaluate
  • 29. ● An A/B test still allows us to evaluate the long term behavior of a given policy Evaluating Online Non-personalized Personalized variant 1 Personalized variant 2 Personalized variant 3
  • 30. Recap ● Making a personalized, timely decision for every Netflix subscriber ● Removing bias from the system ● Maximizing causal impact ● Balancing reward against cost ● Lots of exciting future work….. Check out more in Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Bandits with Large Action Spaces