SlideShare a Scribd company logo
1 of 57
Recent Trends in
Personalization
at Netflix
Justin Basilico
RecSys 2020 Expo
2020-09-24
@JustinBasilico
Why do we personalize?
Help members find content
to watch and enjoy to maximize
member satisfaction and retention
Spark joy
What do we personalize?
Ordering of videos is personalized
From how we rank
Ranking
Selection and placement of rows is personalized
... to how we construct a pageRows
... to how we respond to queries
Search query & result recommendation
... to what images we suggest
Frame recommendation for artists
Personalized artwork selection
... and then select
... to how we reach out
Message personalization
Everything is a recommendation!
Isn’t this solved yet?
○ Every person is unique with a variety of interests
… and sometimes they share profiles
○ Help people find what they want when they’re not sure what they want
○ Large datasets but small data per user
… and potentially biased by the output of your system
○ Cold-start problems on all sides
○ Non-stationary, context-dependent, mood-dependent, ...
○ More than just accuracy: Diversity, novelty, freshness, fairness, ...
○ ...
No, personalization is hard!
So what are you doing about it?
Some recent avenues in approaching these challenges:
1. Causality
2. Bandits
3. Reinforcement Learning
4. Objectives
5. Fairness
6. Experience Personalization
Trending Now
Trend 1: Causality
From Correlation to Causation
● Most recommendation algorithms
are correlational
○ Some early recommendation
algorithms literally computed
correlations between users and items
● Did you watch a movie because
we recommended it to you? Or
because you liked it? Or both?
● If you had to watch a movie, would
you like it? [Wang et al., 2020] p(Y|X) → p(Y|X, do(R))
(from http://www.tylervigen.com/spurious-correlations)
Feedback loops
Impression bias
inflates plays
Leads to inflated
item popularity
More plays
More
impressions
Oscillations in
distribution of genre
recommendations
Feedback loops can cause biases to be
reinforced by the recommendation system!
[Chaney et al., 2018]: simulations showing that this can reduce the
usefulness of the system
Lots of feedback loops...
Closed Loop
Training
Data
Watches Model
Recs
Search
Training
Data
Watches Model
Recs
Open Loop
Closed Loop
Training
Data
Watches Model
Recs
Search
Training
Data
Watches Model
Recs
Open Loop
Challenges in Causal Recommendations
● Handling unobserved confounders
● Coming up with the right causal graph for the model
● High variance in many causal models
● Computational challenges (e.g. [Wong, 2020])
● Connecting causal recommendations with other aspects like
off-policy reinforcement learning
● When and how to introduce randomization
Trend 2: Bandits in
Recommendations
Why contextual bandits for recommendations?
● Break feedback loops
● Want to explore to learn
● Uncertainty around user interests and new items
● Sparse and indirect feedback
● Changing trends
▶Early news example: [Li et al., 2010]
Example:
Which artwork to show?
Artwork Personalization as
Contextual Bandit
● Environment: Netflix homepage
● Context: Member, device, page, etc.
● Learner: Artwork selector for a show
● Action: Display specific image for show
● Reward: Member has positive engagement
Artwork Selector
▶
Offline Replay Results
● Bandit finds good images
● Personalization is better
● Artwork variety matters
● Personalization wiggles
around best images
Lift in Replay in the various algorithms as
compared to the Random baseline
[More info in our blog post]
● Designing good exploration is an art
○ Especially to support future algorithm innovation
○ Challenging to do user-level A/B tests comparing fully
on-policy bandits at high scale
● Bandits over large action spaces: rankings and slates
● Layers of bandits that influence each other
● Handling delayed rewards
Challenges in with bandits in the real world
Trend 3: Reinforcement
Learning in
Recommendations
Going Long-Term
● Want to maximize long-term member joy
● Involves many user visits, recommendation actions and delayed reward
● … sounds like Reinforcement Learning
Within a page
RL to optimize a
ranking or slate
How long?
Within a session
RL to optimize
multiple interactions
in a session
Across sessions
RL to optimize
interactions across
multiple sessions
● High-dimensional: Action of recommending a single item is O(|C|);
typically want to do ranking or page construction, which is combinatorial.
So are states such as user histories.
● Off-policy: Need to learn and evaluate from existing system actions
● Concurrent: Don’t observe full trajectories, need to learn simultaneously
from many interactions
● Evolving action space: New actions (items) become available and need to
be cold-started. Non-stationary behavior for existing actions.
● Simulator paradox: A great simulator means you already have a great
recommender
● Reward function design: Expressing the objective in a good way
Challenges of Reinforcement Learning for
Recommendations
Interested in more?
REVEAL Workshop 2020:
Bandit and Reinforcement Learning from User Interactions
Trend 4: Objectives
● We want to optimize long-term member joy
● While accounting for:
○ Avoiding “trust busters”
○ Coldstarting
○ Fairness
○ ...
What is your recommender trying to optimize?
Layers of Metrics
Training
Objective
Offline Metric Online Metric Goal
Layers of Metrics
RMSE
NDCG on
historical data
User
Engagement in
A/B test
Joy
Example case: Misaligned Metrics
Training
Objective
Offline Metric Online Metric Goal
Your recommendations can only be as good as the
metrics you measure it on
Many recommenders to optimize
● Same objective? Different ones?
● Can we train (some of) them
together using multi-task learning?
● Is there a way to know a-priori if
combining tasks will be beneficial
or not?
User
history
Ranking
Page
Rating
Explanation
Search
Image
Context ...
[Some MTL examples: Zhao et al., 2015, Bansal et al., 2016, Lu et al., 2018, ...]
● Nuanced metrics:
○ Differences between what you want and what you can
encapsulate in a metric
○ Where does enjoyment come from? How does that vary by
person?
○ How do you measure that at scale?
● Ways of measuring improvements offline before going to A/B test?
● What about effects beyond typical A/B time horizon?
● Avoiding introducing lots of parameters to tune
Challenges in objectives
Trend 5: Fairness
Personalization has a big impact in people’s lives
How do we ensure that it is fair?
Calibrated Recommendations [Steck, 2018]
● Fairness as matching distribution of user interests
● Accuracy as an objective can lead to unbalanced predictions
● Simple example:
● Many recommendation algorithms exhibit this behavior of exaggerating the
dominant interests and crowd out less frequent ones
30 action70 romance
30% action70% romance
User:
Expectation:
100% romanceReality: Maximizes accuracy
Calibration Results (MovieLens 20M)
Baseline model (wMF):
Many users receive
uncalibrated rec’s
After reranking:
Rec’s are much more
calibrated (smaller )
Userdensity
More calibrated (KL divergence)
Submodular
Reranker:
● Which definition of fairness to use in different recommendation
scenarios? [Mehrabi et. al, 2019 catalogues many types]
● Handling fairness without demographic information: both
methods [Beutel et al., 2020] and metrics
● Relationship of fairness with explainability and trust
● Connecting Fairness with all the prior areas
○ Bandits, RL, causality, …
● Beyond fairness of the algorithm: ensuring a positive impact on
society
Challenges in fairness for recommenders
Trend 6:
Experience Personalization
Rating Ranking Pages
4.7
Experience
Evolution of our Personalization Approach
Personalizing how we recommend
(not just what we recommend…)
● Algorithm level: Ideal balance of diversity, popularity,
novelty, freshness, etc. may depend on the person
● Display level: How you present items or explain
recommendations can also be personalized
● Interaction level: Balancing the needs of lean-back
users and power users
So many dimensions to personalize
Rows
Trailer
Evidence
Synopsis
Image
Row Title
Metadata
Ranking
More Adaptive UI
Experience beyond the app
Recommendations New Arrival New Season AlertComing Soon
[Slides about messaging]
● Novelty and learning effects for new experiences
● Cohesion across pages, devices, and time
● Dealing with indirect feedback
● Handling structures of components
○ See [Elahi & Chandrashekar, 2020] poster today
● Coldstarting new experiences
Challenges in Experience Personalization
1. Causality
2. Bandits
3. Reinforcement Learning
4. Objectives
5. Fairness
6. Experience Personalization
Lots of opportunities to improve our
Personalization
Sound interesting?Join us
research.netflix.com/jobs
Interested in internship opportunities?
Follow @NetflixResearch
Thank you
Questions?
@JustinBasilico
Justin Basilico

More Related Content

What's hot

Contextualization at Netflix
Contextualization at NetflixContextualization at Netflix
Contextualization at NetflixLinas Baltrunas
 
Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations at NetflixLinas Baltrunas
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsJustin Basilico
 
Artwork Personalization at Netflix
Artwork Personalization at NetflixArtwork Personalization at Netflix
Artwork Personalization at NetflixJustin Basilico
 
Calibrated Recommendations
Calibrated RecommendationsCalibrated Recommendations
Calibrated RecommendationsHarald Steck
 
Missing values in recommender models
Missing values in recommender modelsMissing values in recommender models
Missing values in recommender modelsParmeshwar Khurd
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsYves Raimond
 
Artwork Personalization at Netflix Fernando Amat RecSys2018
Artwork Personalization at Netflix Fernando Amat RecSys2018 Artwork Personalization at Netflix Fernando Amat RecSys2018
Artwork Personalization at Netflix Fernando Amat RecSys2018 Fernando Amat
 
Personalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsPersonalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsJustin Basilico
 
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
 
Sequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsSequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsJaya Kawale
 
Personalization at Netflix - Making Stories Travel
Personalization at Netflix -  Making Stories Travel Personalization at Netflix -  Making Stories Travel
Personalization at Netflix - Making Stories Travel Sudeep Das, Ph.D.
 
Making Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableMaking Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableJustin 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
 
Past, present, and future of Recommender Systems: an industry perspective
Past, present, and future of Recommender Systems: an industry perspectivePast, present, and future of Recommender Systems: an industry perspective
Past, present, and future of Recommender Systems: an industry perspectiveXavier Amatriain
 
Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...
Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...
Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...Xavier Amatriain
 
Recommender system introduction
Recommender system   introductionRecommender system   introduction
Recommender system introductionLiang Xiang
 
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.
 
Recommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareRecommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareJustin Basilico
 

What's hot (20)

Contextualization at Netflix
Contextualization at NetflixContextualization at Netflix
Contextualization at Netflix
 
Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations at Netflix
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Artwork Personalization at Netflix
Artwork Personalization at NetflixArtwork Personalization at Netflix
Artwork Personalization at Netflix
 
Calibrated Recommendations
Calibrated RecommendationsCalibrated Recommendations
Calibrated Recommendations
 
Missing values in recommender models
Missing values in recommender modelsMissing values in recommender models
Missing values in recommender models
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Artwork Personalization at Netflix Fernando Amat RecSys2018
Artwork Personalization at Netflix Fernando Amat RecSys2018 Artwork Personalization at Netflix Fernando Amat RecSys2018
Artwork Personalization at Netflix Fernando Amat RecSys2018
 
Personalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsPersonalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing Recommendations
 
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
 
Sequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsSequential Decision Making in Recommendations
Sequential Decision Making in Recommendations
 
Personalization at Netflix - Making Stories Travel
Personalization at Netflix -  Making Stories Travel Personalization at Netflix -  Making Stories Travel
Personalization at Netflix - Making Stories Travel
 
Making Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableMaking Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms Reliable
 
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
 
Past, present, and future of Recommender Systems: an industry perspective
Past, present, and future of Recommender Systems: an industry perspectivePast, present, and future of Recommender Systems: an industry perspective
Past, present, and future of Recommender Systems: an industry perspective
 
Recent Trends in Personalization at Netflix
Recent Trends in Personalization at NetflixRecent Trends in Personalization at Netflix
Recent Trends in Personalization at Netflix
 
Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...
Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...
Recsys 2016 tutorial: Lessons learned from building real-life recommender sys...
 
Recommender system introduction
Recommender system   introductionRecommender system   introduction
Recommender system introduction
 
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!
 
Recommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareRecommendations for Building Machine Learning Software
Recommendations for Building Machine Learning Software
 

Similar to Recent Trends in Personalization at Netflix

Recommender Systems In Industry
Recommender Systems In IndustryRecommender Systems In Industry
Recommender Systems In IndustryXavier Amatriain
 
Digital analytics lecture1
Digital analytics lecture1Digital analytics lecture1
Digital analytics lecture1Joni Salminen
 
Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...
Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...
Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...Clark Boyd
 
The Hive Think Tank: Machine Learning at Pinterest by Jure Leskovec
The Hive Think Tank: Machine Learning at Pinterest by Jure LeskovecThe Hive Think Tank: Machine Learning at Pinterest by Jure Leskovec
The Hive Think Tank: Machine Learning at Pinterest by Jure LeskovecThe Hive
 
Marketing for NonProfit Organizations
Marketing for NonProfit OrganizationsMarketing for NonProfit Organizations
Marketing for NonProfit OrganizationsAnalytive
 
Leverage The Power of Small Data
Leverage The Power of Small DataLeverage The Power of Small Data
Leverage The Power of Small DataKaryn Zuidinga
 
How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...
How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...
How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...jagan477830
 
Requirements Engineering for the Humanities
Requirements Engineering for the HumanitiesRequirements Engineering for the Humanities
Requirements Engineering for the HumanitiesShawn Day
 
Identifying Personas With Agile Research - Dawn of the Data Age Lecture Series
Identifying Personas With Agile Research - Dawn of the Data Age Lecture SeriesIdentifying Personas With Agile Research - Dawn of the Data Age Lecture Series
Identifying Personas With Agile Research - Dawn of the Data Age Lecture SeriesLuciano Pesci, PhD
 
User Experience Research: Deriving Insights for Customer Development
User Experience Research: Deriving Insights for Customer DevelopmentUser Experience Research: Deriving Insights for Customer Development
User Experience Research: Deriving Insights for Customer DevelopmentNoreen Whysel
 
Week 1 Welcome, intro & overview-1.pptx
Week 1 Welcome, intro & overview-1.pptxWeek 1 Welcome, intro & overview-1.pptx
Week 1 Welcome, intro & overview-1.pptxMingLin41
 
Solving Design and Business Problems in 3 Days with Google Design Sprint by B...
Solving Design and Business Problems in 3 Days with Google Design Sprint by B...Solving Design and Business Problems in 3 Days with Google Design Sprint by B...
Solving Design and Business Problems in 3 Days with Google Design Sprint by B...Borrys Hasian
 
2023-04-11-who-ai-win-fbg.pdf
2023-04-11-who-ai-win-fbg.pdf2023-04-11-who-ai-win-fbg.pdf
2023-04-11-who-ai-win-fbg.pdfJonti Bolles
 
Why your analytics land with a thud
Why your analytics land with a thudWhy your analytics land with a thud
Why your analytics land with a thudOne North
 
How to get stakeholder buy in for ux research
How to get stakeholder buy in for ux researchHow to get stakeholder buy in for ux research
How to get stakeholder buy in for ux researchAlicia Zhong
 
[UPDATE] Udacity webinar on Recommendation Systems
[UPDATE] Udacity webinar on Recommendation Systems[UPDATE] Udacity webinar on Recommendation Systems
[UPDATE] Udacity webinar on Recommendation SystemsAxel de Romblay
 
Aiinpractice2017deepaklongversion
Aiinpractice2017deepaklongversionAiinpractice2017deepaklongversion
Aiinpractice2017deepaklongversionDeepak Agarwal
 
Udacity webinar on Recommendation Systems
Udacity webinar on Recommendation SystemsUdacity webinar on Recommendation Systems
Udacity webinar on Recommendation SystemsAxel de Romblay
 

Similar to Recent Trends in Personalization at Netflix (20)

Recommender Systems In Industry
Recommender Systems In IndustryRecommender Systems In Industry
Recommender Systems In Industry
 
Digital analytics lecture1
Digital analytics lecture1Digital analytics lecture1
Digital analytics lecture1
 
Building a Usability Practice
Building a Usability PracticeBuilding a Usability Practice
Building a Usability Practice
 
Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...
Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...
Your smarter data analytics strategy - Social Media Strategies Summit (SMSS) ...
 
The Hive Think Tank: Machine Learning at Pinterest by Jure Leskovec
The Hive Think Tank: Machine Learning at Pinterest by Jure LeskovecThe Hive Think Tank: Machine Learning at Pinterest by Jure Leskovec
The Hive Think Tank: Machine Learning at Pinterest by Jure Leskovec
 
Marketing for NonProfit Organizations
Marketing for NonProfit OrganizationsMarketing for NonProfit Organizations
Marketing for NonProfit Organizations
 
Leverage The Power of Small Data
Leverage The Power of Small DataLeverage The Power of Small Data
Leverage The Power of Small Data
 
How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...
How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...
How to Choose the Perfect Mtech Project Topic for Your Interests and Career G...
 
Requirements Engineering for the Humanities
Requirements Engineering for the HumanitiesRequirements Engineering for the Humanities
Requirements Engineering for the Humanities
 
Identifying Personas With Agile Research - Dawn of the Data Age Lecture Series
Identifying Personas With Agile Research - Dawn of the Data Age Lecture SeriesIdentifying Personas With Agile Research - Dawn of the Data Age Lecture Series
Identifying Personas With Agile Research - Dawn of the Data Age Lecture Series
 
User Experience Research: Deriving Insights for Customer Development
User Experience Research: Deriving Insights for Customer DevelopmentUser Experience Research: Deriving Insights for Customer Development
User Experience Research: Deriving Insights for Customer Development
 
Week 1 Welcome, intro & overview-1.pptx
Week 1 Welcome, intro & overview-1.pptxWeek 1 Welcome, intro & overview-1.pptx
Week 1 Welcome, intro & overview-1.pptx
 
Solving Design and Business Problems in 3 Days with Google Design Sprint by B...
Solving Design and Business Problems in 3 Days with Google Design Sprint by B...Solving Design and Business Problems in 3 Days with Google Design Sprint by B...
Solving Design and Business Problems in 3 Days with Google Design Sprint by B...
 
2023-04-11-who-ai-win-fbg.pdf
2023-04-11-who-ai-win-fbg.pdf2023-04-11-who-ai-win-fbg.pdf
2023-04-11-who-ai-win-fbg.pdf
 
Why your analytics land with a thud
Why your analytics land with a thudWhy your analytics land with a thud
Why your analytics land with a thud
 
The UX Analyst
The UX AnalystThe UX Analyst
The UX Analyst
 
How to get stakeholder buy in for ux research
How to get stakeholder buy in for ux researchHow to get stakeholder buy in for ux research
How to get stakeholder buy in for ux research
 
[UPDATE] Udacity webinar on Recommendation Systems
[UPDATE] Udacity webinar on Recommendation Systems[UPDATE] Udacity webinar on Recommendation Systems
[UPDATE] Udacity webinar on Recommendation Systems
 
Aiinpractice2017deepaklongversion
Aiinpractice2017deepaklongversionAiinpractice2017deepaklongversion
Aiinpractice2017deepaklongversion
 
Udacity webinar on Recommendation Systems
Udacity webinar on Recommendation SystemsUdacity webinar on Recommendation Systems
Udacity webinar on Recommendation Systems
 

More from Justin Basilico

Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...
Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...
Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...Justin Basilico
 
Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...
Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...
Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...Justin Basilico
 
Recommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareRecommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareJustin Basilico
 
Lessons Learned from Building Machine Learning Software at Netflix
Lessons Learned from Building Machine Learning Software at NetflixLessons Learned from Building Machine Learning Software at Netflix
Lessons Learned from Building Machine Learning Software at NetflixJustin Basilico
 
Recommendation at Netflix Scale
Recommendation at Netflix ScaleRecommendation at Netflix Scale
Recommendation at Netflix ScaleJustin Basilico
 

More from Justin Basilico (6)

Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...
Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...
Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Band...
 
Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...
Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...
Is that a Time Machine? Some Design Patterns for Real World Machine Learning ...
 
Recommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareRecommendations for Building Machine Learning Software
Recommendations for Building Machine Learning Software
 
Lessons Learned from Building Machine Learning Software at Netflix
Lessons Learned from Building Machine Learning Software at NetflixLessons Learned from Building Machine Learning Software at Netflix
Lessons Learned from Building Machine Learning Software at Netflix
 
Learning to Personalize
Learning to PersonalizeLearning to Personalize
Learning to Personalize
 
Recommendation at Netflix Scale
Recommendation at Netflix ScaleRecommendation at Netflix Scale
Recommendation at Netflix Scale
 

Recently uploaded

TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data DiscoveryTrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data DiscoveryTrustArc
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FMESafe Software
 
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamDEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamUiPathCommunity
 
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost SavingRepurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost SavingEdi Saputra
 
presentation ICT roal in 21st century education
presentation ICT roal in 21st century educationpresentation ICT roal in 21st century education
presentation ICT roal in 21st century educationjfdjdjcjdnsjd
 
AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)
AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)
AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)Samir Dash
 
Six Myths about Ontologies: The Basics of Formal Ontology
Six Myths about Ontologies: The Basics of Formal OntologySix Myths about Ontologies: The Basics of Formal Ontology
Six Myths about Ontologies: The Basics of Formal Ontologyjohnbeverley2021
 
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...apidays
 
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...DianaGray10
 
Mcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Mcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot ModelMcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Mcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot ModelDeepika Singh
 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherRemote DBA Services
 
JohnPollard-hybrid-app-RailsConf2024.pptx
JohnPollard-hybrid-app-RailsConf2024.pptxJohnPollard-hybrid-app-RailsConf2024.pptx
JohnPollard-hybrid-app-RailsConf2024.pptxJohnPollard37
 
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...apidays
 
Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxRustici Software
 
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...Orbitshub
 
Finding Java's Hidden Performance Traps @ DevoxxUK 2024
Finding Java's Hidden Performance Traps @ DevoxxUK 2024Finding Java's Hidden Performance Traps @ DevoxxUK 2024
Finding Java's Hidden Performance Traps @ DevoxxUK 2024Victor Rentea
 
Why Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire businessWhy Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire businesspanagenda
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWERMadyBayot
 
CNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In PakistanCNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In Pakistandanishmna97
 
Polkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin WoodPolkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin WoodJuan lago vázquez
 

Recently uploaded (20)

TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data DiscoveryTrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
 
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamDEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
 
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost SavingRepurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
 
presentation ICT roal in 21st century education
presentation ICT roal in 21st century educationpresentation ICT roal in 21st century education
presentation ICT roal in 21st century education
 
AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)
AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)
AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)
 
Six Myths about Ontologies: The Basics of Formal Ontology
Six Myths about Ontologies: The Basics of Formal OntologySix Myths about Ontologies: The Basics of Formal Ontology
Six Myths about Ontologies: The Basics of Formal Ontology
 
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
 
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
 
Mcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Mcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot ModelMcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Mcleodganj Call Girls 🥰 8617370543 Service Offer VIP Hot Model
 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a Fresher
 
JohnPollard-hybrid-app-RailsConf2024.pptx
JohnPollard-hybrid-app-RailsConf2024.pptxJohnPollard-hybrid-app-RailsConf2024.pptx
JohnPollard-hybrid-app-RailsConf2024.pptx
 
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
 
Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptx
 
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
 
Finding Java's Hidden Performance Traps @ DevoxxUK 2024
Finding Java's Hidden Performance Traps @ DevoxxUK 2024Finding Java's Hidden Performance Traps @ DevoxxUK 2024
Finding Java's Hidden Performance Traps @ DevoxxUK 2024
 
Why Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire businessWhy Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire business
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
 
CNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In PakistanCNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In Pakistan
 
Polkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin WoodPolkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin Wood
 

Recent Trends in Personalization at Netflix

  • 1. Recent Trends in Personalization at Netflix Justin Basilico RecSys 2020 Expo 2020-09-24 @JustinBasilico
  • 2. Why do we personalize?
  • 3. Help members find content to watch and enjoy to maximize member satisfaction and retention
  • 5. What do we personalize?
  • 6. Ordering of videos is personalized From how we rank Ranking
  • 7. Selection and placement of rows is personalized ... to how we construct a pageRows
  • 8. ... to how we respond to queries Search query & result recommendation
  • 9. ... to what images we suggest Frame recommendation for artists
  • 11. ... to how we reach out Message personalization
  • 12. Everything is a recommendation!
  • 14. ○ Every person is unique with a variety of interests … and sometimes they share profiles ○ Help people find what they want when they’re not sure what they want ○ Large datasets but small data per user … and potentially biased by the output of your system ○ Cold-start problems on all sides ○ Non-stationary, context-dependent, mood-dependent, ... ○ More than just accuracy: Diversity, novelty, freshness, fairness, ... ○ ... No, personalization is hard!
  • 15. So what are you doing about it?
  • 16. Some recent avenues in approaching these challenges: 1. Causality 2. Bandits 3. Reinforcement Learning 4. Objectives 5. Fairness 6. Experience Personalization Trending Now
  • 18. From Correlation to Causation ● Most recommendation algorithms are correlational ○ Some early recommendation algorithms literally computed correlations between users and items ● Did you watch a movie because we recommended it to you? Or because you liked it? Or both? ● If you had to watch a movie, would you like it? [Wang et al., 2020] p(Y|X) → p(Y|X, do(R)) (from http://www.tylervigen.com/spurious-correlations)
  • 19. Feedback loops Impression bias inflates plays Leads to inflated item popularity More plays More impressions Oscillations in distribution of genre recommendations Feedback loops can cause biases to be reinforced by the recommendation system! [Chaney et al., 2018]: simulations showing that this can reduce the usefulness of the system
  • 20. Lots of feedback loops...
  • 23. Challenges in Causal Recommendations ● Handling unobserved confounders ● Coming up with the right causal graph for the model ● High variance in many causal models ● Computational challenges (e.g. [Wong, 2020]) ● Connecting causal recommendations with other aspects like off-policy reinforcement learning ● When and how to introduce randomization
  • 24. Trend 2: Bandits in Recommendations
  • 25. Why contextual bandits for recommendations? ● Break feedback loops ● Want to explore to learn ● Uncertainty around user interests and new items ● Sparse and indirect feedback ● Changing trends ▶Early news example: [Li et al., 2010]
  • 27. Artwork Personalization as Contextual Bandit ● Environment: Netflix homepage ● Context: Member, device, page, etc. ● Learner: Artwork selector for a show ● Action: Display specific image for show ● Reward: Member has positive engagement Artwork Selector ▶
  • 28. Offline Replay Results ● Bandit finds good images ● Personalization is better ● Artwork variety matters ● Personalization wiggles around best images Lift in Replay in the various algorithms as compared to the Random baseline [More info in our blog post]
  • 29. ● Designing good exploration is an art ○ Especially to support future algorithm innovation ○ Challenging to do user-level A/B tests comparing fully on-policy bandits at high scale ● Bandits over large action spaces: rankings and slates ● Layers of bandits that influence each other ● Handling delayed rewards Challenges in with bandits in the real world
  • 30. Trend 3: Reinforcement Learning in Recommendations
  • 31. Going Long-Term ● Want to maximize long-term member joy ● Involves many user visits, recommendation actions and delayed reward ● … sounds like Reinforcement Learning
  • 32. Within a page RL to optimize a ranking or slate How long? Within a session RL to optimize multiple interactions in a session Across sessions RL to optimize interactions across multiple sessions
  • 33. ● High-dimensional: Action of recommending a single item is O(|C|); typically want to do ranking or page construction, which is combinatorial. So are states such as user histories. ● Off-policy: Need to learn and evaluate from existing system actions ● Concurrent: Don’t observe full trajectories, need to learn simultaneously from many interactions ● Evolving action space: New actions (items) become available and need to be cold-started. Non-stationary behavior for existing actions. ● Simulator paradox: A great simulator means you already have a great recommender ● Reward function design: Expressing the objective in a good way Challenges of Reinforcement Learning for Recommendations
  • 34. Interested in more? REVEAL Workshop 2020: Bandit and Reinforcement Learning from User Interactions
  • 36. ● We want to optimize long-term member joy ● While accounting for: ○ Avoiding “trust busters” ○ Coldstarting ○ Fairness ○ ... What is your recommender trying to optimize?
  • 37. Layers of Metrics Training Objective Offline Metric Online Metric Goal
  • 38. Layers of Metrics RMSE NDCG on historical data User Engagement in A/B test Joy Example case: Misaligned Metrics Training Objective Offline Metric Online Metric Goal
  • 39. Your recommendations can only be as good as the metrics you measure it on
  • 40. Many recommenders to optimize ● Same objective? Different ones? ● Can we train (some of) them together using multi-task learning? ● Is there a way to know a-priori if combining tasks will be beneficial or not? User history Ranking Page Rating Explanation Search Image Context ... [Some MTL examples: Zhao et al., 2015, Bansal et al., 2016, Lu et al., 2018, ...]
  • 41. ● Nuanced metrics: ○ Differences between what you want and what you can encapsulate in a metric ○ Where does enjoyment come from? How does that vary by person? ○ How do you measure that at scale? ● Ways of measuring improvements offline before going to A/B test? ● What about effects beyond typical A/B time horizon? ● Avoiding introducing lots of parameters to tune Challenges in objectives
  • 43. Personalization has a big impact in people’s lives How do we ensure that it is fair?
  • 44. Calibrated Recommendations [Steck, 2018] ● Fairness as matching distribution of user interests ● Accuracy as an objective can lead to unbalanced predictions ● Simple example: ● Many recommendation algorithms exhibit this behavior of exaggerating the dominant interests and crowd out less frequent ones 30 action70 romance 30% action70% romance User: Expectation: 100% romanceReality: Maximizes accuracy
  • 45. Calibration Results (MovieLens 20M) Baseline model (wMF): Many users receive uncalibrated rec’s After reranking: Rec’s are much more calibrated (smaller ) Userdensity More calibrated (KL divergence) Submodular Reranker:
  • 46. ● Which definition of fairness to use in different recommendation scenarios? [Mehrabi et. al, 2019 catalogues many types] ● Handling fairness without demographic information: both methods [Beutel et al., 2020] and metrics ● Relationship of fairness with explainability and trust ● Connecting Fairness with all the prior areas ○ Bandits, RL, causality, … ● Beyond fairness of the algorithm: ensuring a positive impact on society Challenges in fairness for recommenders
  • 48. Rating Ranking Pages 4.7 Experience Evolution of our Personalization Approach
  • 49. Personalizing how we recommend (not just what we recommend…) ● Algorithm level: Ideal balance of diversity, popularity, novelty, freshness, etc. may depend on the person ● Display level: How you present items or explain recommendations can also be personalized ● Interaction level: Balancing the needs of lean-back users and power users
  • 50. So many dimensions to personalize Rows Trailer Evidence Synopsis Image Row Title Metadata Ranking
  • 52. Experience beyond the app Recommendations New Arrival New Season AlertComing Soon [Slides about messaging]
  • 53. ● Novelty and learning effects for new experiences ● Cohesion across pages, devices, and time ● Dealing with indirect feedback ● Handling structures of components ○ See [Elahi & Chandrashekar, 2020] poster today ● Coldstarting new experiences Challenges in Experience Personalization
  • 54.
  • 55. 1. Causality 2. Bandits 3. Reinforcement Learning 4. Objectives 5. Fairness 6. Experience Personalization Lots of opportunities to improve our Personalization
  • 56. Sound interesting?Join us research.netflix.com/jobs Interested in internship opportunities? Follow @NetflixResearch