SlideShare a Scribd company logo
1 of 22
Download to read offline
© Tubi, proprietary and confidential
Exploration and Diversity
in Recommender Systems
Jaya Kawale
Director of Machine Learning
2020/08/13
© Tubi, proprietary and confidential 2
Tubi
● AVOD (Advertising based
Video On Demand)
● Acquired by Fox.
● Surpassed 200M hours
viewed per month in June.
© Tubi, proprietary and confidential 3
ML at Tubi
Important areas
● Personalization ML
● Content ML
● Ads ML
ML for Personalization
● Homepage recommendations
● Container/rows ordering
● Search
● Notifications
1st & 3rd
Party Data
Audience
Assessment
Viewer-oriented data
Universe of Content + Metadata
Title-orienteddata
Products
Models
Embeddings (CTXT, MD, MMD,
Genre, Demos, Actor, et al)
Use Cases
Beam from
Universe to
Tubiverse
Cold➔
Warm➔Hot
Starting
Content Value
Assessment
Tiering
Inventory in
Tubiverse
Augmented
Search
Seeding
Growth
Coordinated
Pursuit of New
Audience
Portfolio
Analysis /
Simulation
ML for Content
© Tubi, proprietary and confidential 6
Core Idea: Recommend items based
upon the previous interactions of the
users.
Popular approaches: Collaborative
Filtering.
Recommender system
Buy
People
Also Buy
Recommend
© Tubi, proprietary and confidential
Exploration
© Tubi, proprietary and confidential 8
Production bias: Only get feedback from recommended items. Only show titles
with sufficient feedback.
Why do we need Exploration ?
Problem:
● How to cold start new titles/users ?
● How to handle changing user
tastes ?
● How to handle changing title
popularities ?
Recommend
ML Model
© Tubi, proprietary and confidential 9
Exploration-Exploitation Tradeoff: Explore unknown
items to gather feedback or exploit current best item
Exploration in Recommendation
© Tubi, proprietary and confidential 10
P(Epsilon): Explore
P(1- Epsilon): Exploit
Epsilon-Greedy
© Tubi, proprietary and confidential 11
Optimism in the Face of Uncertainty
Principle: “Choose your actions as if the
environment is as nice as plausible”
Optimism leads to exploration - reward from
the unknown arm possibly greater
Upper Confidence Bound
Arm 1 Arm 2
X
X
X
X
X
X
X
Mu 1 Mu 2
UCB1
UCB2
Optimistic estimate = Mu + CI
© Tubi, proprietary and confidential 12
Maintain a posterior distribution over the
parameters for each action.
Sample the estimated reward
Recommend the item with the highest
estimated reward
Update the posterior.
Thompson Sampling
Arm1
Arm2
Arm3
Arm4
© Tubi, proprietary and confidential 13
No assumptions on a reward distribution.
Reward non-stationary.
Based upon Exponential Weighted algorithm “Hedge”
Algorithms slower in practice.
Adversarial setting Exp3
© Tubi, proprietary and confidential 14
Additional context provided to the learner
No single arm is good in all cases
Classical MAB: Compete against the
best arm
Contextual MAB: Compete against the
best “policy”
Contextual Bandits
Context
Past Viewing History
Time of Day
Country/DMA
Recommend
© Tubi, proprietary and confidential 15
Many practical challenges
● Arms keep changing
● Non-stationary rewards
● Budget constraints
● Slate of recommendations
● Feedback loops
Bandits in the real world
© Tubi, proprietary and confidential
Diversity
© Tubi, proprietary and confidential 17
Correlations: Ordering items by relevance score could result in correlations
between nearby items
Why do we need Diversity?
Problems:
● Utility of the “set” of recommendations
not maximized.
● Real estate better utilized by other title
© Tubi, proprietary and confidential 18
“Remove redundant items to replace with alternatives that the user will enjoy
concurrently.”
Compute a diversity score
Methods based on coverage and redundancy
Diversity in Recommendations
rerankScore = (1 - lambda) * relevancyScore + lambda * diversityScore
© Tubi, proprietary and confidential 19
YouTube homepage diversification
DPP - characterized by the determinant of some function.
Let Relevancy score = P1, P2, … PN; Similarity score = Sij
Determinantal Point Process
P1 S12 S13 S14
S12 P2 S23 S24
S13 S23 P3 S34
S14 S24 S34 P4
Determinant
© Tubi, proprietary and confidential 20
Exploration
Primary Goal: To understand user
preferences for items as there is
imperfect information.
Exploration is still needed for
uncorrelated recommendations.
The story of Exploration and Diversity
Diversity
Primary Goal: Get rid of correlated
recommendations for better user
experience.
Diversity is still needed when we know
the user preferences accurately.
Exploration can be designed to achieve diversification and diversification can help achieve exploration!
© Tubi, proprietary and confidential 21
Lattimore, T., & Szepesvári, C. (2020). Bandit Algorithms. Cambridge: Cambridge
University Press. doi:10.1017/9781108571401
Wilhelm, Mark & Ramanathan, Ajith & Bonomo, Alexander & Jain, Sagar & Chi, Ed
& Gillenwater, Jennifer. (2018). Practical Diversified Recommendations on
YouTube with Determinantal Point Processes. 2165-2173.
10.1145/3269206.3272018.
References
© Tubi, proprietary and confidential 22
Thank You!
We are hiring!
Email: jkawale@tubi.tv
Twitter: jayakawale

More Related Content

What's hot

Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations at NetflixLinas Baltrunas
 
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
 
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
 
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
 
Missing values in recommender models
Missing values in recommender modelsMissing values in recommender models
Missing values in recommender modelsParmeshwar Khurd
 
Sequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsSequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsJaya Kawale
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsYves Raimond
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsJustin Basilico
 
Recent Trends in Personalization at Netflix
Recent Trends in Personalization at NetflixRecent Trends in Personalization at Netflix
Recent Trends in Personalization at NetflixJustin Basilico
 
Recommendation Modeling with Impression Data at Netflix
Recommendation Modeling with Impression Data at NetflixRecommendation Modeling with Impression Data at Netflix
Recommendation Modeling with Impression Data at NetflixJiangwei Pan
 
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
 
Recommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareRecommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareJustin Basilico
 
Contextualization at Netflix
Contextualization at NetflixContextualization at Netflix
Contextualization at NetflixLinas Baltrunas
 
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
 
Data council SF 2020 Building a Personalized Messaging System at Netflix
Data council SF 2020 Building a Personalized Messaging System at NetflixData council SF 2020 Building a Personalized Messaging System at Netflix
Data council SF 2020 Building a Personalized Messaging System at NetflixGrace T. Huang
 
Personalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsPersonalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsJustin Basilico
 
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
 
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
 
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.
 

What's hot (20)

Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations at Netflix
 
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
 
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
 
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
 
Missing values in recommender models
Missing values in recommender modelsMissing values in recommender models
Missing values in recommender models
 
Sequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsSequential Decision Making in Recommendations
Sequential Decision Making in Recommendations
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Recent Trends in Personalization at Netflix
Recent Trends in Personalization at NetflixRecent Trends in Personalization at Netflix
Recent Trends in Personalization at Netflix
 
Recommendation Modeling with Impression Data at Netflix
Recommendation Modeling with Impression Data at NetflixRecommendation Modeling with Impression Data at Netflix
Recommendation Modeling with Impression Data at Netflix
 
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
 
Recommendations for Building Machine Learning Software
Recommendations for Building Machine Learning SoftwareRecommendations for Building Machine Learning Software
Recommendations for Building Machine Learning Software
 
Contextualization at Netflix
Contextualization at NetflixContextualization at Netflix
Contextualization at Netflix
 
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
 
Data council SF 2020 Building a Personalized Messaging System at Netflix
Data council SF 2020 Building a Personalized Messaging System at NetflixData council SF 2020 Building a Personalized Messaging System at Netflix
Data council SF 2020 Building a Personalized Messaging System at Netflix
 
Personalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing RecommendationsPersonalized Page Generation for Browsing Recommendations
Personalized Page Generation for Browsing Recommendations
 
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
 
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
 
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!
 

Similar to Exploration and diversity in recommender systems

Jaya WWW talk 2023.pdf
Jaya WWW talk 2023.pdfJaya WWW talk 2023.pdf
Jaya WWW talk 2023.pdfJaya Kawale
 
Machine Learning At Tubi
Machine Learning At TubiMachine Learning At Tubi
Machine Learning At TubiJaya Kawale
 
Udacity webinar on Recommendation Systems
Udacity webinar on Recommendation SystemsUdacity webinar on Recommendation Systems
Udacity webinar on Recommendation SystemsAxel de Romblay
 
[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
 
2017 10-10 (netflix ml platform meetup) learning item and user representation...
2017 10-10 (netflix ml platform meetup) learning item and user representation...2017 10-10 (netflix ml platform meetup) learning item and user representation...
2017 10-10 (netflix ml platform meetup) learning item and user representation...Ed Chi
 
[UMAP 2016] User-Oriented Context Suggestion
[UMAP 2016] User-Oriented Context Suggestion[UMAP 2016] User-Oriented Context Suggestion
[UMAP 2016] User-Oriented Context SuggestionYONG ZHENG
 
A flexible recommenndation system for Cable TV
A flexible recommenndation system for Cable TVA flexible recommenndation system for Cable TV
A flexible recommenndation system for Cable TVIntoTheMinds
 
A Flexible Recommendation System for Cable TV
A Flexible Recommendation System for Cable TVA Flexible Recommendation System for Cable TV
A Flexible Recommendation System for Cable TVFrancisco Couto
 
105 Advanced A-B Testing: Making Decisions with Data
105 Advanced A-B Testing: Making Decisions with Data105 Advanced A-B Testing: Making Decisions with Data
105 Advanced A-B Testing: Making Decisions with DataProductCamp Boston
 
The Economics of Data Sharing
The Economics of Data SharingThe Economics of Data Sharing
The Economics of Data SharingAnita de Waard
 
recommendation system techunique and issue
recommendation system techunique and issuerecommendation system techunique and issue
recommendation system techunique and issueNutanBhor
 
A_B Testing Personalized Meditation Recommendations.pdf
A_B Testing Personalized Meditation Recommendations.pdfA_B Testing Personalized Meditation Recommendations.pdf
A_B Testing Personalized Meditation Recommendations.pdfVWO
 
AI Driven Product Innovation
AI Driven Product InnovationAI Driven Product Innovation
AI Driven Product Innovationebelani
 
AI-driven product innovation: from Recommender Systems to COVID-19
AI-driven product innovation: from Recommender Systems to COVID-19AI-driven product innovation: from Recommender Systems to COVID-19
AI-driven product innovation: from Recommender Systems to COVID-19Xavier Amatriain
 
Data driven approaches in a technology startup
Data driven approaches in a technology startupData driven approaches in a technology startup
Data driven approaches in a technology startupRakuten Group, Inc.
 
Persuasive Message Evaluation with Google AdWords
Persuasive Message Evaluation with Google AdWordsPersuasive Message Evaluation with Google AdWords
Persuasive Message Evaluation with Google AdWordsMarco Guerini
 
Recommandation systems -
Recommandation systems - Recommandation systems -
Recommandation systems - Yousef Fadila
 
Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...
Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...
Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...SigOpt
 
Predictive Solutions and Analytics for TV & Entertainment Businesses
Predictive Solutions and Analytics for TV & Entertainment BusinessesPredictive Solutions and Analytics for TV & Entertainment Businesses
Predictive Solutions and Analytics for TV & Entertainment BusinessesDavid Zibriczky
 

Similar to Exploration and diversity in recommender systems (20)

Jaya WWW talk 2023.pdf
Jaya WWW talk 2023.pdfJaya WWW talk 2023.pdf
Jaya WWW talk 2023.pdf
 
Machine Learning At Tubi
Machine Learning At TubiMachine Learning At Tubi
Machine Learning At Tubi
 
Udacity webinar on Recommendation Systems
Udacity webinar on Recommendation SystemsUdacity webinar on Recommendation Systems
Udacity webinar on Recommendation Systems
 
[UPDATE] Udacity webinar on Recommendation Systems
[UPDATE] Udacity webinar on Recommendation Systems[UPDATE] Udacity webinar on Recommendation Systems
[UPDATE] Udacity webinar on Recommendation Systems
 
2017 10-10 (netflix ml platform meetup) learning item and user representation...
2017 10-10 (netflix ml platform meetup) learning item and user representation...2017 10-10 (netflix ml platform meetup) learning item and user representation...
2017 10-10 (netflix ml platform meetup) learning item and user representation...
 
[UMAP 2016] User-Oriented Context Suggestion
[UMAP 2016] User-Oriented Context Suggestion[UMAP 2016] User-Oriented Context Suggestion
[UMAP 2016] User-Oriented Context Suggestion
 
A flexible recommenndation system for Cable TV
A flexible recommenndation system for Cable TVA flexible recommenndation system for Cable TV
A flexible recommenndation system for Cable TV
 
A Flexible Recommendation System for Cable TV
A Flexible Recommendation System for Cable TVA Flexible Recommendation System for Cable TV
A Flexible Recommendation System for Cable TV
 
105 Advanced A-B Testing: Making Decisions with Data
105 Advanced A-B Testing: Making Decisions with Data105 Advanced A-B Testing: Making Decisions with Data
105 Advanced A-B Testing: Making Decisions with Data
 
The Economics of Data Sharing
The Economics of Data SharingThe Economics of Data Sharing
The Economics of Data Sharing
 
recommendation system techunique and issue
recommendation system techunique and issuerecommendation system techunique and issue
recommendation system techunique and issue
 
A_B Testing Personalized Meditation Recommendations.pdf
A_B Testing Personalized Meditation Recommendations.pdfA_B Testing Personalized Meditation Recommendations.pdf
A_B Testing Personalized Meditation Recommendations.pdf
 
AI Driven Product Innovation
AI Driven Product InnovationAI Driven Product Innovation
AI Driven Product Innovation
 
AI-driven product innovation: from Recommender Systems to COVID-19
AI-driven product innovation: from Recommender Systems to COVID-19AI-driven product innovation: from Recommender Systems to COVID-19
AI-driven product innovation: from Recommender Systems to COVID-19
 
Data driven approaches in a technology startup
Data driven approaches in a technology startupData driven approaches in a technology startup
Data driven approaches in a technology startup
 
Persuasive Message Evaluation with Google AdWords
Persuasive Message Evaluation with Google AdWordsPersuasive Message Evaluation with Google AdWords
Persuasive Message Evaluation with Google AdWords
 
Eval rec algo_crowdsourcing__icalt_2014_ma
Eval rec algo_crowdsourcing__icalt_2014_maEval rec algo_crowdsourcing__icalt_2014_ma
Eval rec algo_crowdsourcing__icalt_2014_ma
 
Recommandation systems -
Recommandation systems - Recommandation systems -
Recommandation systems -
 
Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...
Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...
Interactive Tradeoffs Between Competing Offline Metrics with Bayesian Optimiz...
 
Predictive Solutions and Analytics for TV & Entertainment Businesses
Predictive Solutions and Analytics for TV & Entertainment BusinessesPredictive Solutions and Analytics for TV & Entertainment Businesses
Predictive Solutions and Analytics for TV & Entertainment Businesses
 

Recently uploaded

Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)dollysharma2066
 
Solving The Right Triangles PowerPoint 2.ppt
Solving The Right Triangles PowerPoint 2.pptSolving The Right Triangles PowerPoint 2.ppt
Solving The Right Triangles PowerPoint 2.pptJasonTagapanGulla
 
Why does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsync
Why does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsyncWhy does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsync
Why does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsyncssuser2ae721
 
Class 1 | NFPA 72 | Overview Fire Alarm System
Class 1 | NFPA 72 | Overview Fire Alarm SystemClass 1 | NFPA 72 | Overview Fire Alarm System
Class 1 | NFPA 72 | Overview Fire Alarm Systemirfanmechengr
 
CCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdf
CCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdfCCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdf
CCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdfAsst.prof M.Gokilavani
 
Work Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvvWork Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvvLewisJB
 
Piping Basic stress analysis by engineering
Piping Basic stress analysis by engineeringPiping Basic stress analysis by engineering
Piping Basic stress analysis by engineeringJuanCarlosMorales19600
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.eptoze12
 
Transport layer issues and challenges - Guide
Transport layer issues and challenges - GuideTransport layer issues and challenges - Guide
Transport layer issues and challenges - GuideGOPINATHS437943
 
Sachpazis Costas: Geotechnical Engineering: A student's Perspective Introduction
Sachpazis Costas: Geotechnical Engineering: A student's Perspective IntroductionSachpazis Costas: Geotechnical Engineering: A student's Perspective Introduction
Sachpazis Costas: Geotechnical Engineering: A student's Perspective IntroductionDr.Costas Sachpazis
 
TechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor Catchers
TechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor CatchersTechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor Catchers
TechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor Catcherssdickerson1
 
Correctly Loading Incremental Data at Scale
Correctly Loading Incremental Data at ScaleCorrectly Loading Incremental Data at Scale
Correctly Loading Incremental Data at ScaleAlluxio, Inc.
 
Past, Present and Future of Generative AI
Past, Present and Future of Generative AIPast, Present and Future of Generative AI
Past, Present and Future of Generative AIabhishek36461
 
welding defects observed during the welding
welding defects observed during the weldingwelding defects observed during the welding
welding defects observed during the weldingMuhammadUzairLiaqat
 
Concrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptxConcrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptxKartikeyaDwivedi3
 
IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024Mark Billinghurst
 
An experimental study in using natural admixture as an alternative for chemic...
An experimental study in using natural admixture as an alternative for chemic...An experimental study in using natural admixture as an alternative for chemic...
An experimental study in using natural admixture as an alternative for chemic...Chandu841456
 

Recently uploaded (20)

Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
 
Solving The Right Triangles PowerPoint 2.ppt
Solving The Right Triangles PowerPoint 2.pptSolving The Right Triangles PowerPoint 2.ppt
Solving The Right Triangles PowerPoint 2.ppt
 
Why does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsync
Why does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsyncWhy does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsync
Why does (not) Kafka need fsync: Eliminating tail latency spikes caused by fsync
 
POWER SYSTEMS-1 Complete notes examples
POWER SYSTEMS-1 Complete notes  examplesPOWER SYSTEMS-1 Complete notes  examples
POWER SYSTEMS-1 Complete notes examples
 
Class 1 | NFPA 72 | Overview Fire Alarm System
Class 1 | NFPA 72 | Overview Fire Alarm SystemClass 1 | NFPA 72 | Overview Fire Alarm System
Class 1 | NFPA 72 | Overview Fire Alarm System
 
CCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdf
CCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdfCCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdf
CCS355 Neural Networks & Deep Learning Unit 1 PDF notes with Question bank .pdf
 
Design and analysis of solar grass cutter.pdf
Design and analysis of solar grass cutter.pdfDesign and analysis of solar grass cutter.pdf
Design and analysis of solar grass cutter.pdf
 
Work Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvvWork Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvv
 
Piping Basic stress analysis by engineering
Piping Basic stress analysis by engineeringPiping Basic stress analysis by engineering
Piping Basic stress analysis by engineering
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.
 
Transport layer issues and challenges - Guide
Transport layer issues and challenges - GuideTransport layer issues and challenges - Guide
Transport layer issues and challenges - Guide
 
Sachpazis Costas: Geotechnical Engineering: A student's Perspective Introduction
Sachpazis Costas: Geotechnical Engineering: A student's Perspective IntroductionSachpazis Costas: Geotechnical Engineering: A student's Perspective Introduction
Sachpazis Costas: Geotechnical Engineering: A student's Perspective Introduction
 
TechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor Catchers
TechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor CatchersTechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor Catchers
TechTAC® CFD Report Summary: A Comparison of Two Types of Tubing Anchor Catchers
 
Correctly Loading Incremental Data at Scale
Correctly Loading Incremental Data at ScaleCorrectly Loading Incremental Data at Scale
Correctly Loading Incremental Data at Scale
 
Past, Present and Future of Generative AI
Past, Present and Future of Generative AIPast, Present and Future of Generative AI
Past, Present and Future of Generative AI
 
welding defects observed during the welding
welding defects observed during the weldingwelding defects observed during the welding
welding defects observed during the welding
 
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Serviceyoung call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
 
Concrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptxConcrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptx
 
IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024
 
An experimental study in using natural admixture as an alternative for chemic...
An experimental study in using natural admixture as an alternative for chemic...An experimental study in using natural admixture as an alternative for chemic...
An experimental study in using natural admixture as an alternative for chemic...
 

Exploration and diversity in recommender systems

  • 1. © Tubi, proprietary and confidential Exploration and Diversity in Recommender Systems Jaya Kawale Director of Machine Learning 2020/08/13
  • 2. © Tubi, proprietary and confidential 2 Tubi ● AVOD (Advertising based Video On Demand) ● Acquired by Fox. ● Surpassed 200M hours viewed per month in June.
  • 3. © Tubi, proprietary and confidential 3 ML at Tubi Important areas ● Personalization ML ● Content ML ● Ads ML
  • 4. ML for Personalization ● Homepage recommendations ● Container/rows ordering ● Search ● Notifications
  • 5. 1st & 3rd Party Data Audience Assessment Viewer-oriented data Universe of Content + Metadata Title-orienteddata Products Models Embeddings (CTXT, MD, MMD, Genre, Demos, Actor, et al) Use Cases Beam from Universe to Tubiverse Cold➔ Warm➔Hot Starting Content Value Assessment Tiering Inventory in Tubiverse Augmented Search Seeding Growth Coordinated Pursuit of New Audience Portfolio Analysis / Simulation ML for Content
  • 6. © Tubi, proprietary and confidential 6 Core Idea: Recommend items based upon the previous interactions of the users. Popular approaches: Collaborative Filtering. Recommender system Buy People Also Buy Recommend
  • 7. © Tubi, proprietary and confidential Exploration
  • 8. © Tubi, proprietary and confidential 8 Production bias: Only get feedback from recommended items. Only show titles with sufficient feedback. Why do we need Exploration ? Problem: ● How to cold start new titles/users ? ● How to handle changing user tastes ? ● How to handle changing title popularities ? Recommend ML Model
  • 9. © Tubi, proprietary and confidential 9 Exploration-Exploitation Tradeoff: Explore unknown items to gather feedback or exploit current best item Exploration in Recommendation
  • 10. © Tubi, proprietary and confidential 10 P(Epsilon): Explore P(1- Epsilon): Exploit Epsilon-Greedy
  • 11. © Tubi, proprietary and confidential 11 Optimism in the Face of Uncertainty Principle: “Choose your actions as if the environment is as nice as plausible” Optimism leads to exploration - reward from the unknown arm possibly greater Upper Confidence Bound Arm 1 Arm 2 X X X X X X X Mu 1 Mu 2 UCB1 UCB2 Optimistic estimate = Mu + CI
  • 12. © Tubi, proprietary and confidential 12 Maintain a posterior distribution over the parameters for each action. Sample the estimated reward Recommend the item with the highest estimated reward Update the posterior. Thompson Sampling Arm1 Arm2 Arm3 Arm4
  • 13. © Tubi, proprietary and confidential 13 No assumptions on a reward distribution. Reward non-stationary. Based upon Exponential Weighted algorithm “Hedge” Algorithms slower in practice. Adversarial setting Exp3
  • 14. © Tubi, proprietary and confidential 14 Additional context provided to the learner No single arm is good in all cases Classical MAB: Compete against the best arm Contextual MAB: Compete against the best “policy” Contextual Bandits Context Past Viewing History Time of Day Country/DMA Recommend
  • 15. © Tubi, proprietary and confidential 15 Many practical challenges ● Arms keep changing ● Non-stationary rewards ● Budget constraints ● Slate of recommendations ● Feedback loops Bandits in the real world
  • 16. © Tubi, proprietary and confidential Diversity
  • 17. © Tubi, proprietary and confidential 17 Correlations: Ordering items by relevance score could result in correlations between nearby items Why do we need Diversity? Problems: ● Utility of the “set” of recommendations not maximized. ● Real estate better utilized by other title
  • 18. © Tubi, proprietary and confidential 18 “Remove redundant items to replace with alternatives that the user will enjoy concurrently.” Compute a diversity score Methods based on coverage and redundancy Diversity in Recommendations rerankScore = (1 - lambda) * relevancyScore + lambda * diversityScore
  • 19. © Tubi, proprietary and confidential 19 YouTube homepage diversification DPP - characterized by the determinant of some function. Let Relevancy score = P1, P2, … PN; Similarity score = Sij Determinantal Point Process P1 S12 S13 S14 S12 P2 S23 S24 S13 S23 P3 S34 S14 S24 S34 P4 Determinant
  • 20. © Tubi, proprietary and confidential 20 Exploration Primary Goal: To understand user preferences for items as there is imperfect information. Exploration is still needed for uncorrelated recommendations. The story of Exploration and Diversity Diversity Primary Goal: Get rid of correlated recommendations for better user experience. Diversity is still needed when we know the user preferences accurately. Exploration can be designed to achieve diversification and diversification can help achieve exploration!
  • 21. © Tubi, proprietary and confidential 21 Lattimore, T., & Szepesvári, C. (2020). Bandit Algorithms. Cambridge: Cambridge University Press. doi:10.1017/9781108571401 Wilhelm, Mark & Ramanathan, Ajith & Bonomo, Alexander & Jain, Sagar & Chi, Ed & Gillenwater, Jennifer. (2018). Practical Diversified Recommendations on YouTube with Determinantal Point Processes. 2165-2173. 10.1145/3269206.3272018. References
  • 22. © Tubi, proprietary and confidential 22 Thank You! We are hiring! Email: jkawale@tubi.tv Twitter: jayakawale