SlideShare a Scribd company logo
1 of 35
Download to read offline
Neighbor methods vs.
matrix factorization:
case studies of real-life recommendations
By István Pilászy, András Serény, György Dózsa,
Balázs Hidasi, Attila Sári, János Gub
Presented by:
István Pilászy <pila@gravityrd.com>
at RecSys’15
About Gravity R&D
9/22/2015
• Founded by 4 people participating in the Netflix Prize
competition, tied for the first place
• VC Investment in 2009 and 2011
• Providing Recommendation as a Service
• Clients in 6 continents
About the Netflix Prize
9/22/2015
• 2006-2009
• Predict movie ratings (explicit feedback)
• CBF did not work
• Classical CF methods (item-kNN, user-kNN) did not
work
• Matrix Factorization was extremely good
• We were fully in love with Matrix Factorization
Case study #1:
Netflix Prize demo
Netflix Prize demo / 1
9/22/2015
• In 2009 we created a public demo mainly for investors
• Users can rate movies and get recommendations
• What do you expect from a demo?
 Be relevant even after 1 rating
 Users will provide their favorite movies first
 Be relevant after 2 ratings: both movies should affect the
results
Netflix Prize demo / 2
9/22/2015
• Using a good MF model with K=200 factors and biases
• Use linear regression to compute user feature vector
• Recs after rating a romantic movie Notting Hill, 1999
OK Score Title
 4.6916 The_Shawshank_Redemption/1994
 4.6858 House,_M.D.:_Season_1/2004
 4.6825 Lost:_Season_1/2004
 4.5903 Anne_of_Green_Gables:_The_Sequel/1987
 4.5497 Lord_of_the_Rings:_The_Return_of_the_King/2003
Netflix Prize demo / 3
9/22/2015
• Idea: turn off item bias during recommendation.
• Result are fully relevant
• Even with 10 factors, it is very good
OK Score Title
 4.3323 Love_Actually/2003
 4.3015 Runaway_Bride/1999
 4.2811 My_Best_Friend's_Wedding/1997
 4.2790 You've_Got_Mail/1998
 4.1564 About_a_Boy/2002
Netflix Prize demo / 4
9/22/2015
• Now give 5-star rating to Saving Private Ryan / 1998
• Almost no change in the list
OK Score Title
 4.5911 You've_Got_Mail/1998
 4.5085 Love_Actually/2003
 4.3944 Sleepless_in_Seattle/1993
 4.3625 Runaway_Bride/1999
 4.3274 My_Best_Friend's_Wedding/1997
Netflix Prize demo / 5
9/22/2015
• Idea: set item biases to zero before computing user feature vector
• 5th rec is romantic + war
• Conclusion: MF is good, but rating and ranking are very different
OK Score Title
 4.5094 You've_Got_Mail/1998
 4.3445 Black_Hawk_Down/2001
 4.3298 Sleepless_in_Seattle/1993
 4.3114 Love_Actually/2003
! 4.2805 Apollo_13/1995
About real life problems
About real life problems: item2item recs
9/22/2015
• What is item-to-item recommendation?
 People who viewed this also viewed: …
 Viewed, watched, bought, liked, favorited, etc.
• Ignoring the current user
• The recommendation should be relevant to the current
item
• Very common scenario
9/22/2015
About real life problems / 2
9/22/2015
• In Academic papers
 50% explicit feedback
 50% implicit feedback
o 49.9% personal
o 0.1% item2item
• At gravityrd.com:
 1% explicit feedback
 99% implicit feedback
o 15% personal
o 84% item2item
• Sites where rating is crucial tend to create their own rec engine
• Even if there is explicit rating, there are more implicit feedback
About real life problems / 3
9/22/2015
• Data characteristics (after data retention):
 Number of active users: 100k – 100M
 Number of active items : 1k – 100M
 Number of relations between them: 10M – 10B
• Response time: must be within 200ms
• We cannot give 199ms for MF prediction + 1ms
business logic
Time complexity of MF for implicit feedback
9/22/2015
• During training
 E = #events, U = #users, I = #items
 implicit ALS: E*K^2 + U*K^3 + I*K^3
o with Coordinate Descent: E*K + U*K^2 + I*K^2
o with CG: the same, but more stable.
 BPR: E * K
 CliMF: O(E * K * avg(user support))
• During recommendation: I * K
• Not practical if I > 100k, K > 100
• You have to increase K as I grows
Case study #2:
item-to-item recommendations
with SVD
i2i recommendations with SVD / 2
9/22/2015
• Recommendations should seem relevant
• You can expect that movies of the same trilogy are similar to each
other (tetralogies too)
• We defined the following metric:
 For movies A and B of a trilogy, check if B is amongst the top-5 most
similar items of A.
Score: 0 or 1
 A trilogy can provide 6 such pairs (12 for tetralogies)
 Sum up this for all trilogies
• We used a custom movie dataset
• Good metric for CF item-to-item, bad metric for CBF item-to-item
i2i recommendations with SVD / 3
9/22/2015
• Evaluating for SVD with different number of factors
• Using cosine similarity between SVD feature vectors
• more factors provide better results
• Why not use the original space?
• Who wants to run SVD with 500 factors?
• Score of neighbor method (using cosine similarity between
original vectors): 169
K 10 20 50 100 200 500 1000 1500
score 72 82 95 96 106 126 152 158
I2i recommendations with SVD / 4
9/22/2015
• What does a 200-factor SVD recommend to
Kill Bill: Vol. 1
• Really bad recommendation
OK Cos
Sim
Title
 0.299 Kill Bill: Vol. 2
 0.273 Matthias, Matthias
 0.223 The New Rijksmuseum
 0.199 Naked
 0.190 Grave Danger
i2i recommendations with SVD / 5
9/22/2015
• What does a 1500-factor SVD recommend to
Kill Bill: Vol. 1
• Good, but uses lots of CPU
• But that is an easy domain, with 20k movies!
OK Cos
Sim
Title
 0.292 Kill Bill: Vol. 2
! 0.140 Inglourious Basterds
! 0.133 Pulp Fiction
 0.131 American Beauty
! 0.125 Reservoir Dogs
Case Study #3:
Implementing an item-to-item
method
Implementing an item-to-item method / 1
9/22/2015
We implemented the following article:
Noam Koenigstein and Yehuda Koren. "Towards scalable and
accurate item-oriented recommendations." Proceedings of the 7th
ACM conference on Recommender systems. ACM, 2013.
• They define a new metric for i2i evaluation:
MPR (Mean Percentile Rank):
If user visits A, and then B, then recommend for A, and see the
position of B in that list.
• They propose a new method (EIR, Euclidean Item Recommender) ,
that assigns feature vector for each item, so that if A is close to B,
then users frequently visit B after A.
• They don’t compare it with pure popularity method
Implementing an item-to-item method / 2
9/22/2015
Results on a custom movie dataset:
• SVD and other methods can’t beat the new method
• Popularity method is better or on-pair with the new method
• Recommendations for Pulp Fiction:
SVD New method
Reservoir Dogs A Space Odyssey
Inglourious Basterds A Clockwork Orange
Four Rooms The Godfather
The Shawshank Redemption Eternal Sunshine of the Spotless Mind
Fight Club Mulholland Drive
Implementing an item-to-item method / 3
9/22/2015
Comparison
method
metadata similarity
(larger is better)
MPR
(smaller is better)
cos 7.54 0.68
jaccard 7.59 0.68
Assoc.rule 6.44 0.68
pop 1.65 0.25
random 1.44 0.50
EIR 5.00 0.25
Implementing an item-to-item method / 4
9/22/2015
Summary of this method:
• This method is better in MPR than many other methods
• It is on pair with Popularity method
• It is worse in metadata-based similarity
• Sometimes recommendations look like they were random
• Sensitive to the parameters
• Very few articles are dealing with CF item-to-item recs
Case studies on CTR
Case studies on CTR / 1
9/22/2015
CTR almost doubled when we switched from IALS1 to
item-kNN on a site where users and items are the same
9/22/2015
Case studies on CTR / 2
9/22/2015
Comparison of BPR vs. item-kNN on a classified site, for
item-to-item recommendations.
Item-kNN is the winner
9/22/2015
Item-kNN
BPR
Case studies on CTR / 3
9/22/2015
Using BPR vs. item-kNN on a video site for personal
recommendations.
Measuring number of clicks on recommendations.
Result: 4% more clicks for BPR.
9/22/2015
BPR
Item-kNN
Case studies on CTR / 4
9/22/2015
Conclusions / 1
9/22/2015
Problems with MF:
• Lots of parameters to tune
• Needs many iteration over the data
• if there is no inter-connection between two item sets,
they can get similar feature vectors.
• Sensitive to noise in data (e.g. users with few events)
• Not the best for item-to-item recs, especially when
many neighbors exist already
Conclusions / 2
9/22/2015
When to use Matrix Factorization
• you have one, dense domain (e.g. movies), with not too many items
(e.g. less than 100k)
• feedback is taste-based
• For personalized recommendations (e.g. newsletter)
• try both with A/B testing
• try blending
 Must be smart in blending (e.g. using it for high supported items)
• Usually better for offline evaluation metrics

More Related Content

What's hot

What's hot (20)

Business Intelligence
Business IntelligenceBusiness Intelligence
Business Intelligence
 
Market basket analysis
Market basket analysisMarket basket analysis
Market basket analysis
 
Data Visualization With Tableau | Edureka
Data Visualization With Tableau | EdurekaData Visualization With Tableau | Edureka
Data Visualization With Tableau | Edureka
 
Data Mining: Association Rules Basics
Data Mining: Association Rules BasicsData Mining: Association Rules Basics
Data Mining: Association Rules Basics
 
Business Intelligence - A Management Perspective
Business Intelligence - A Management PerspectiveBusiness Intelligence - A Management Perspective
Business Intelligence - A Management Perspective
 
Apriori algorithm
Apriori algorithmApriori algorithm
Apriori algorithm
 
Handling noisy data
Handling noisy dataHandling noisy data
Handling noisy data
 
Data warehouse and olap technology
Data warehouse and olap technologyData warehouse and olap technology
Data warehouse and olap technology
 
DATA WAREHOUSE IMPLEMENTATION BY SAIKIRAN PANJALA
DATA WAREHOUSE IMPLEMENTATION BY SAIKIRAN PANJALADATA WAREHOUSE IMPLEMENTATION BY SAIKIRAN PANJALA
DATA WAREHOUSE IMPLEMENTATION BY SAIKIRAN PANJALA
 
Business intelligence concepts & application
Business intelligence concepts & applicationBusiness intelligence concepts & application
Business intelligence concepts & application
 
Apriori
AprioriApriori
Apriori
 
Data visualisation & analytics with Tableau
Data visualisation & analytics with Tableau Data visualisation & analytics with Tableau
Data visualisation & analytics with Tableau
 
Chapter 4. Data Warehousing and On-Line Analytical Processing.ppt
Chapter 4. Data Warehousing and On-Line Analytical Processing.pptChapter 4. Data Warehousing and On-Line Analytical Processing.ppt
Chapter 4. Data Warehousing and On-Line Analytical Processing.ppt
 
Association rule mining
Association rule miningAssociation rule mining
Association rule mining
 
07. Analytics & Reporting Requirements Template
07. Analytics & Reporting Requirements Template07. Analytics & Reporting Requirements Template
07. Analytics & Reporting Requirements Template
 
Data Visualization
Data VisualizationData Visualization
Data Visualization
 
Business Intelligence Presentation (1/2)
Business Intelligence Presentation (1/2)Business Intelligence Presentation (1/2)
Business Intelligence Presentation (1/2)
 
Understanding Association Rule Mining
Understanding Association Rule MiningUnderstanding Association Rule Mining
Understanding Association Rule Mining
 
Brief introduction to data visualization
Brief introduction to data visualizationBrief introduction to data visualization
Brief introduction to data visualization
 
Business intelligence
Business intelligenceBusiness intelligence
Business intelligence
 

Viewers also liked

Factorization Machines with libFM
Factorization Machines with libFMFactorization Machines with libFM
Factorization Machines with libFM
Liangjie Hong
 
Matrix factorization
Matrix factorizationMatrix factorization
Matrix factorization
rubyyc
 
Recommender system introduction
Recommender system   introductionRecommender system   introduction
Recommender system introduction
Liang Xiang
 
Lecture 6 lu factorization & determinants - section 2-5 2-7 3-1 and 3-2
Lecture 6   lu factorization & determinants - section 2-5 2-7 3-1 and 3-2Lecture 6   lu factorization & determinants - section 2-5 2-7 3-1 and 3-2
Lecture 6 lu factorization & determinants - section 2-5 2-7 3-1 and 3-2
njit-ronbrown
 
Building a Recommendation Engine - An example of a product recommendation engine
Building a Recommendation Engine - An example of a product recommendation engineBuilding a Recommendation Engine - An example of a product recommendation engine
Building a Recommendation Engine - An example of a product recommendation engine
NYC Predictive Analytics
 
Recommender system algorithm and architecture
Recommender system algorithm and architectureRecommender system algorithm and architecture
Recommender system algorithm and architecture
Liang Xiang
 

Viewers also liked (17)

Nonnegative Matrix Factorization
Nonnegative Matrix FactorizationNonnegative Matrix Factorization
Nonnegative Matrix Factorization
 
Factorization Machines with libFM
Factorization Machines with libFMFactorization Machines with libFM
Factorization Machines with libFM
 
Matrix factorization
Matrix factorizationMatrix factorization
Matrix factorization
 
Collaborative Filtering with Spark
Collaborative Filtering with SparkCollaborative Filtering with Spark
Collaborative Filtering with Spark
 
Intro to Factorization Machines
Intro to Factorization MachinesIntro to Factorization Machines
Intro to Factorization Machines
 
آموزش محاسبات عددی - بخش دوم
آموزش محاسبات عددی - بخش دومآموزش محاسبات عددی - بخش دوم
آموزش محاسبات عددی - بخش دوم
 
Matrix Factorization Technique for Recommender Systems
Matrix Factorization Technique for Recommender SystemsMatrix Factorization Technique for Recommender Systems
Matrix Factorization Technique for Recommender Systems
 
Recommender system introduction
Recommender system   introductionRecommender system   introduction
Recommender system introduction
 
Lecture 6 lu factorization & determinants - section 2-5 2-7 3-1 and 3-2
Lecture 6   lu factorization & determinants - section 2-5 2-7 3-1 and 3-2Lecture 6   lu factorization & determinants - section 2-5 2-7 3-1 and 3-2
Lecture 6 lu factorization & determinants - section 2-5 2-7 3-1 and 3-2
 
Introduction to Matrix Factorization Methods Collaborative Filtering
Introduction to Matrix Factorization Methods Collaborative FilteringIntroduction to Matrix Factorization Methods Collaborative Filtering
Introduction to Matrix Factorization Methods Collaborative Filtering
 
Recommender Systems
Recommender SystemsRecommender Systems
Recommender Systems
 
Beginners Guide to Non-Negative Matrix Factorization
Beginners Guide to Non-Negative Matrix FactorizationBeginners Guide to Non-Negative Matrix Factorization
Beginners Guide to Non-Negative Matrix Factorization
 
Recommender Systems
Recommender SystemsRecommender Systems
Recommender Systems
 
Recommendation system
Recommendation system Recommendation system
Recommendation system
 
Collaborative Filtering Recommendation System
Collaborative Filtering Recommendation SystemCollaborative Filtering Recommendation System
Collaborative Filtering Recommendation System
 
Building a Recommendation Engine - An example of a product recommendation engine
Building a Recommendation Engine - An example of a product recommendation engineBuilding a Recommendation Engine - An example of a product recommendation engine
Building a Recommendation Engine - An example of a product recommendation engine
 
Recommender system algorithm and architecture
Recommender system algorithm and architectureRecommender system algorithm and architecture
Recommender system algorithm and architecture
 

Similar to Neighbor methods vs matrix factorization - case studies of real-life recommendations (Gravity LSRS2015 RECSYS 2015)

sr project poster 8
sr project poster 8sr project poster 8
sr project poster 8
Derek Scott
 
Recommender Systems! @ASAI 2011
Recommender Systems! @ASAI 2011Recommender Systems! @ASAI 2011
Recommender Systems! @ASAI 2011
Ernesto Mislej
 
Netflix Presentation - George's
Netflix Presentation - George'sNetflix Presentation - George's
Netflix Presentation - George's
George Ferko
 
How To Use Benchmarking To Evaluate Performance And Maximize Revenue
How To Use Benchmarking To Evaluate Performance And Maximize RevenueHow To Use Benchmarking To Evaluate Performance And Maximize Revenue
How To Use Benchmarking To Evaluate Performance And Maximize Revenue
GoMio.com
 

Similar to Neighbor methods vs matrix factorization - case studies of real-life recommendations (Gravity LSRS2015 RECSYS 2015) (20)

Lessons learnt at building recommendation services at industry scale
Lessons learnt at building recommendation services at industry scaleLessons learnt at building recommendation services at industry scale
Lessons learnt at building recommendation services at industry scale
 
IPTV vs NETFLIX
IPTV vs NETFLIXIPTV vs NETFLIX
IPTV vs NETFLIX
 
Context-aware Recommendation: A Quick View
Context-aware Recommendation: A Quick ViewContext-aware Recommendation: A Quick View
Context-aware Recommendation: A Quick View
 
Kdd 2014 Tutorial - the recommender problem revisited
Kdd 2014 Tutorial -  the recommender problem revisitedKdd 2014 Tutorial -  the recommender problem revisited
Kdd 2014 Tutorial - the recommender problem revisited
 
How to use data to make a hit tv show
How to use data to make a hit tv showHow to use data to make a hit tv show
How to use data to make a hit tv show
 
Movie Recommendation System - MovieLens Dataset
Movie Recommendation System - MovieLens DatasetMovie Recommendation System - MovieLens Dataset
Movie Recommendation System - MovieLens Dataset
 
Movie lens movie recommendation system
Movie lens movie recommendation systemMovie lens movie recommendation system
Movie lens movie recommendation system
 
Netflix's Pricing Increase
Netflix's Pricing IncreaseNetflix's Pricing Increase
Netflix's Pricing Increase
 
Content Recommendation through Semantic Annotation of User Reviews and Linked...
Content Recommendation through Semantic Annotation of User Reviews and Linked...Content Recommendation through Semantic Annotation of User Reviews and Linked...
Content Recommendation through Semantic Annotation of User Reviews and Linked...
 
Applications of Machine Learning
Applications of Machine LearningApplications of Machine Learning
Applications of Machine Learning
 
BSSML17 - Introduction, Models, Evaluations
BSSML17 - Introduction, Models, EvaluationsBSSML17 - Introduction, Models, Evaluations
BSSML17 - Introduction, Models, Evaluations
 
Exploration and diversity in recommender systems
Exploration and diversity in recommender systemsExploration and diversity in recommender systems
Exploration and diversity in recommender systems
 
sr project poster 8
sr project poster 8sr project poster 8
sr project poster 8
 
At Your Service: What Netflix and Assessments Have In Common | SoGoSurvey
At Your Service: What Netflix and Assessments Have In Common | SoGoSurveyAt Your Service: What Netflix and Assessments Have In Common | SoGoSurvey
At Your Service: What Netflix and Assessments Have In Common | SoGoSurvey
 
Recommender Systems! @ASAI 2011
Recommender Systems! @ASAI 2011Recommender Systems! @ASAI 2011
Recommender Systems! @ASAI 2011
 
Netflix Presentation - George's
Netflix Presentation - George'sNetflix Presentation - George's
Netflix Presentation - George's
 
How To Use Benchmarking To Evaluate Performance And Maximize Revenue
How To Use Benchmarking To Evaluate Performance And Maximize RevenueHow To Use Benchmarking To Evaluate Performance And Maximize Revenue
How To Use Benchmarking To Evaluate Performance And Maximize Revenue
 
Modern Perspectives on Recommender Systems and their Applications in Mendeley
Modern Perspectives on Recommender Systems and their Applications in MendeleyModern Perspectives on Recommender Systems and their Applications in Mendeley
Modern Perspectives on Recommender Systems and their Applications in Mendeley
 
Btv cover and lighting guide
Btv cover and lighting guideBtv cover and lighting guide
Btv cover and lighting guide
 
Understanding Project Supply Networks
Understanding Project Supply NetworksUnderstanding Project Supply Networks
Understanding Project Supply Networks
 

More from Domonkos Tikk

General factorization framework for context-aware recommendations
General factorization framework for context-aware recommendationsGeneral factorization framework for context-aware recommendations
General factorization framework for context-aware recommendations
Domonkos Tikk
 
Tartalomgazdagítás (content enrichment)
Tartalomgazdagítás (content enrichment) Tartalomgazdagítás (content enrichment)
Tartalomgazdagítás (content enrichment)
Domonkos Tikk
 
Context-aware similarities within the factorization framework - presented at ...
Context-aware similarities within the factorization framework - presented at ...Context-aware similarities within the factorization framework - presented at ...
Context-aware similarities within the factorization framework - presented at ...
Domonkos Tikk
 
Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...
Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...
Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...
Domonkos Tikk
 
Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...
Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...
Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...
Domonkos Tikk
 

More from Domonkos Tikk (11)

Recommenders on video sharing portals - business and algorithmic aspects
Recommenders on video sharing portals - business and algorithmic aspectsRecommenders on video sharing portals - business and algorithmic aspects
Recommenders on video sharing portals - business and algorithmic aspects
 
Challenges Encountered by Scaling Up Recommendation Services at Gravity R&D
Challenges Encountered by Scaling Up Recommendation Services at Gravity R&DChallenges Encountered by Scaling Up Recommendation Services at Gravity R&D
Challenges Encountered by Scaling Up Recommendation Services at Gravity R&D
 
General factorization framework for context-aware recommendations
General factorization framework for context-aware recommendationsGeneral factorization framework for context-aware recommendations
General factorization framework for context-aware recommendations
 
Tartalomgazdagítás (content enrichment)
Tartalomgazdagítás (content enrichment) Tartalomgazdagítás (content enrichment)
Tartalomgazdagítás (content enrichment)
 
Idomaar crowd rec_reference_fw
Idomaar crowd rec_reference_fwIdomaar crowd rec_reference_fw
Idomaar crowd rec_reference_fw
 
Big Data in Online Classifieds
Big Data in Online ClassifiedsBig Data in Online Classifieds
Big Data in Online Classifieds
 
Context-aware similarities within the factorization framework - presented at ...
Context-aware similarities within the factorization framework - presented at ...Context-aware similarities within the factorization framework - presented at ...
Context-aware similarities within the factorization framework - presented at ...
 
Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...
Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...
Slides from CARR 2012 WS - Enhancing Matrix Factorization Through Initializat...
 
Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...
Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...
Fast ALS-Based Tensor Factorization for Context-Aware Recommendation from Imp...
 
Recommender Systems Evaluation: A 3D Benchmark - presented at RUE 2012 worksh...
Recommender Systems Evaluation: A 3D Benchmark - presented at RUE 2012 worksh...Recommender Systems Evaluation: A 3D Benchmark - presented at RUE 2012 worksh...
Recommender Systems Evaluation: A 3D Benchmark - presented at RUE 2012 worksh...
 
From a toolkit of recommendation algorithms into a real business: the Gravity...
From a toolkit of recommendation algorithms into a real business: the Gravity...From a toolkit of recommendation algorithms into a real business: the Gravity...
From a toolkit of recommendation algorithms into a real business: the Gravity...
 

Recently uploaded

Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵
Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵
Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵
Chandigarh Call girls 9053900678 Call girls in Chandigarh
 
( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...
( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...
( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...
nilamkumrai
 
➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men 🔝mehsana🔝 Escorts...
➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men  🔝mehsana🔝   Escorts...➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men  🔝mehsana🔝   Escorts...
➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men 🔝mehsana🔝 Escorts...
nirzagarg
 
Lucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRL
Lucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRLLucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRL
Lucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRL
imonikaupta
 
Call Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort ServiceCall Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
9953056974 Low Rate Call Girls In Saket, Delhi NCR
 
VIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 Booking
dharasingh5698
 

Recently uploaded (20)

Dubai=Desi Dubai Call Girls O525547819 Outdoor Call Girls Dubai
Dubai=Desi Dubai Call Girls O525547819 Outdoor Call Girls DubaiDubai=Desi Dubai Call Girls O525547819 Outdoor Call Girls Dubai
Dubai=Desi Dubai Call Girls O525547819 Outdoor Call Girls Dubai
 
Busty Desi⚡Call Girls in Vasundhara Ghaziabad >༒8448380779 Escort Service
Busty Desi⚡Call Girls in Vasundhara Ghaziabad >༒8448380779 Escort ServiceBusty Desi⚡Call Girls in Vasundhara Ghaziabad >༒8448380779 Escort Service
Busty Desi⚡Call Girls in Vasundhara Ghaziabad >༒8448380779 Escort Service
 
Pirangut | Call Girls Pune Phone No 8005736733 Elite Escort Service Available...
Pirangut | Call Girls Pune Phone No 8005736733 Elite Escort Service Available...Pirangut | Call Girls Pune Phone No 8005736733 Elite Escort Service Available...
Pirangut | Call Girls Pune Phone No 8005736733 Elite Escort Service Available...
 
Yerawada ] Independent Escorts in Pune - Book 8005736733 Call Girls Available...
Yerawada ] Independent Escorts in Pune - Book 8005736733 Call Girls Available...Yerawada ] Independent Escorts in Pune - Book 8005736733 Call Girls Available...
Yerawada ] Independent Escorts in Pune - Book 8005736733 Call Girls Available...
 
(INDIRA) Call Girl Pune Call Now 8250077686 Pune Escorts 24x7
(INDIRA) Call Girl Pune Call Now 8250077686 Pune Escorts 24x7(INDIRA) Call Girl Pune Call Now 8250077686 Pune Escorts 24x7
(INDIRA) Call Girl Pune Call Now 8250077686 Pune Escorts 24x7
 
Katraj ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready For S...
Katraj ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready For S...Katraj ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready For S...
Katraj ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready For S...
 
Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵
Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵
Low Sexy Call Girls In Mohali 9053900678 🥵Have Save And Good Place 🥵
 
( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...
( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...
( Pune ) VIP Baner Call Girls 🎗️ 9352988975 Sizzling | Escorts | Girls Are Re...
 
VVIP Pune Call Girls Sinhagad WhatSapp Number 8005736733 With Elite Staff And...
VVIP Pune Call Girls Sinhagad WhatSapp Number 8005736733 With Elite Staff And...VVIP Pune Call Girls Sinhagad WhatSapp Number 8005736733 With Elite Staff And...
VVIP Pune Call Girls Sinhagad WhatSapp Number 8005736733 With Elite Staff And...
 
Real Escorts in Al Nahda +971524965298 Dubai Escorts Service
Real Escorts in Al Nahda +971524965298 Dubai Escorts ServiceReal Escorts in Al Nahda +971524965298 Dubai Escorts Service
Real Escorts in Al Nahda +971524965298 Dubai Escorts Service
 
➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men 🔝mehsana🔝 Escorts...
➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men  🔝mehsana🔝   Escorts...➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men  🔝mehsana🔝   Escorts...
➥🔝 7737669865 🔝▻ mehsana Call-girls in Women Seeking Men 🔝mehsana🔝 Escorts...
 
20240509 QFM015 Engineering Leadership Reading List April 2024.pdf
20240509 QFM015 Engineering Leadership Reading List April 2024.pdf20240509 QFM015 Engineering Leadership Reading List April 2024.pdf
20240509 QFM015 Engineering Leadership Reading List April 2024.pdf
 
20240510 QFM016 Irresponsible AI Reading List April 2024.pdf
20240510 QFM016 Irresponsible AI Reading List April 2024.pdf20240510 QFM016 Irresponsible AI Reading List April 2024.pdf
20240510 QFM016 Irresponsible AI Reading List April 2024.pdf
 
Lucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRL
Lucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRLLucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRL
Lucknow ❤CALL GIRL 88759*99948 ❤CALL GIRLS IN Lucknow ESCORT SERVICE❤CALL GIRL
 
Call Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort ServiceCall Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Prashant Vihar, Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
 
"Boost Your Digital Presence: Partner with a Leading SEO Agency"
"Boost Your Digital Presence: Partner with a Leading SEO Agency""Boost Your Digital Presence: Partner with a Leading SEO Agency"
"Boost Your Digital Presence: Partner with a Leading SEO Agency"
 
Ganeshkhind ! Call Girls Pune - 450+ Call Girl Cash Payment 8005736733 Neha T...
Ganeshkhind ! Call Girls Pune - 450+ Call Girl Cash Payment 8005736733 Neha T...Ganeshkhind ! Call Girls Pune - 450+ Call Girl Cash Payment 8005736733 Neha T...
Ganeshkhind ! Call Girls Pune - 450+ Call Girl Cash Payment 8005736733 Neha T...
 
Call Now ☎ 8264348440 !! Call Girls in Green Park Escort Service Delhi N.C.R.
Call Now ☎ 8264348440 !! Call Girls in Green Park Escort Service Delhi N.C.R.Call Now ☎ 8264348440 !! Call Girls in Green Park Escort Service Delhi N.C.R.
Call Now ☎ 8264348440 !! Call Girls in Green Park Escort Service Delhi N.C.R.
 
VIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Pollachi 7001035870 Whatsapp Number, 24/07 Booking
 
Shikrapur - Call Girls in Pune Neha 8005736733 | 100% Gennuine High Class Ind...
Shikrapur - Call Girls in Pune Neha 8005736733 | 100% Gennuine High Class Ind...Shikrapur - Call Girls in Pune Neha 8005736733 | 100% Gennuine High Class Ind...
Shikrapur - Call Girls in Pune Neha 8005736733 | 100% Gennuine High Class Ind...
 

Neighbor methods vs matrix factorization - case studies of real-life recommendations (Gravity LSRS2015 RECSYS 2015)

  • 1. Neighbor methods vs. matrix factorization: case studies of real-life recommendations By István Pilászy, András Serény, György Dózsa, Balázs Hidasi, Attila Sári, János Gub Presented by: István Pilászy <pila@gravityrd.com> at RecSys’15
  • 2. About Gravity R&D 9/22/2015 • Founded by 4 people participating in the Netflix Prize competition, tied for the first place • VC Investment in 2009 and 2011 • Providing Recommendation as a Service • Clients in 6 continents
  • 3. About the Netflix Prize 9/22/2015 • 2006-2009 • Predict movie ratings (explicit feedback) • CBF did not work • Classical CF methods (item-kNN, user-kNN) did not work • Matrix Factorization was extremely good • We were fully in love with Matrix Factorization
  • 5. Netflix Prize demo / 1 9/22/2015 • In 2009 we created a public demo mainly for investors • Users can rate movies and get recommendations • What do you expect from a demo?  Be relevant even after 1 rating  Users will provide their favorite movies first  Be relevant after 2 ratings: both movies should affect the results
  • 6. Netflix Prize demo / 2 9/22/2015 • Using a good MF model with K=200 factors and biases • Use linear regression to compute user feature vector • Recs after rating a romantic movie Notting Hill, 1999 OK Score Title  4.6916 The_Shawshank_Redemption/1994  4.6858 House,_M.D.:_Season_1/2004  4.6825 Lost:_Season_1/2004  4.5903 Anne_of_Green_Gables:_The_Sequel/1987  4.5497 Lord_of_the_Rings:_The_Return_of_the_King/2003
  • 7. Netflix Prize demo / 3 9/22/2015 • Idea: turn off item bias during recommendation. • Result are fully relevant • Even with 10 factors, it is very good OK Score Title  4.3323 Love_Actually/2003  4.3015 Runaway_Bride/1999  4.2811 My_Best_Friend's_Wedding/1997  4.2790 You've_Got_Mail/1998  4.1564 About_a_Boy/2002
  • 8. Netflix Prize demo / 4 9/22/2015 • Now give 5-star rating to Saving Private Ryan / 1998 • Almost no change in the list OK Score Title  4.5911 You've_Got_Mail/1998  4.5085 Love_Actually/2003  4.3944 Sleepless_in_Seattle/1993  4.3625 Runaway_Bride/1999  4.3274 My_Best_Friend's_Wedding/1997
  • 9. Netflix Prize demo / 5 9/22/2015 • Idea: set item biases to zero before computing user feature vector • 5th rec is romantic + war • Conclusion: MF is good, but rating and ranking are very different OK Score Title  4.5094 You've_Got_Mail/1998  4.3445 Black_Hawk_Down/2001  4.3298 Sleepless_in_Seattle/1993  4.3114 Love_Actually/2003 ! 4.2805 Apollo_13/1995
  • 10. About real life problems
  • 11. About real life problems: item2item recs 9/22/2015 • What is item-to-item recommendation?  People who viewed this also viewed: …  Viewed, watched, bought, liked, favorited, etc. • Ignoring the current user • The recommendation should be relevant to the current item • Very common scenario
  • 13. About real life problems / 2 9/22/2015 • In Academic papers  50% explicit feedback  50% implicit feedback o 49.9% personal o 0.1% item2item • At gravityrd.com:  1% explicit feedback  99% implicit feedback o 15% personal o 84% item2item • Sites where rating is crucial tend to create their own rec engine • Even if there is explicit rating, there are more implicit feedback
  • 14. About real life problems / 3 9/22/2015 • Data characteristics (after data retention):  Number of active users: 100k – 100M  Number of active items : 1k – 100M  Number of relations between them: 10M – 10B • Response time: must be within 200ms • We cannot give 199ms for MF prediction + 1ms business logic
  • 15. Time complexity of MF for implicit feedback 9/22/2015 • During training  E = #events, U = #users, I = #items  implicit ALS: E*K^2 + U*K^3 + I*K^3 o with Coordinate Descent: E*K + U*K^2 + I*K^2 o with CG: the same, but more stable.  BPR: E * K  CliMF: O(E * K * avg(user support)) • During recommendation: I * K • Not practical if I > 100k, K > 100 • You have to increase K as I grows
  • 16. Case study #2: item-to-item recommendations with SVD
  • 17. i2i recommendations with SVD / 2 9/22/2015 • Recommendations should seem relevant • You can expect that movies of the same trilogy are similar to each other (tetralogies too) • We defined the following metric:  For movies A and B of a trilogy, check if B is amongst the top-5 most similar items of A. Score: 0 or 1  A trilogy can provide 6 such pairs (12 for tetralogies)  Sum up this for all trilogies • We used a custom movie dataset • Good metric for CF item-to-item, bad metric for CBF item-to-item
  • 18. i2i recommendations with SVD / 3 9/22/2015 • Evaluating for SVD with different number of factors • Using cosine similarity between SVD feature vectors • more factors provide better results • Why not use the original space? • Who wants to run SVD with 500 factors? • Score of neighbor method (using cosine similarity between original vectors): 169 K 10 20 50 100 200 500 1000 1500 score 72 82 95 96 106 126 152 158
  • 19. I2i recommendations with SVD / 4 9/22/2015 • What does a 200-factor SVD recommend to Kill Bill: Vol. 1 • Really bad recommendation OK Cos Sim Title  0.299 Kill Bill: Vol. 2  0.273 Matthias, Matthias  0.223 The New Rijksmuseum  0.199 Naked  0.190 Grave Danger
  • 20. i2i recommendations with SVD / 5 9/22/2015 • What does a 1500-factor SVD recommend to Kill Bill: Vol. 1 • Good, but uses lots of CPU • But that is an easy domain, with 20k movies! OK Cos Sim Title  0.292 Kill Bill: Vol. 2 ! 0.140 Inglourious Basterds ! 0.133 Pulp Fiction  0.131 American Beauty ! 0.125 Reservoir Dogs
  • 21. Case Study #3: Implementing an item-to-item method
  • 22. Implementing an item-to-item method / 1 9/22/2015 We implemented the following article: Noam Koenigstein and Yehuda Koren. "Towards scalable and accurate item-oriented recommendations." Proceedings of the 7th ACM conference on Recommender systems. ACM, 2013. • They define a new metric for i2i evaluation: MPR (Mean Percentile Rank): If user visits A, and then B, then recommend for A, and see the position of B in that list. • They propose a new method (EIR, Euclidean Item Recommender) , that assigns feature vector for each item, so that if A is close to B, then users frequently visit B after A. • They don’t compare it with pure popularity method
  • 23. Implementing an item-to-item method / 2 9/22/2015 Results on a custom movie dataset: • SVD and other methods can’t beat the new method • Popularity method is better or on-pair with the new method • Recommendations for Pulp Fiction: SVD New method Reservoir Dogs A Space Odyssey Inglourious Basterds A Clockwork Orange Four Rooms The Godfather The Shawshank Redemption Eternal Sunshine of the Spotless Mind Fight Club Mulholland Drive
  • 24. Implementing an item-to-item method / 3 9/22/2015 Comparison method metadata similarity (larger is better) MPR (smaller is better) cos 7.54 0.68 jaccard 7.59 0.68 Assoc.rule 6.44 0.68 pop 1.65 0.25 random 1.44 0.50 EIR 5.00 0.25
  • 25. Implementing an item-to-item method / 4 9/22/2015 Summary of this method: • This method is better in MPR than many other methods • It is on pair with Popularity method • It is worse in metadata-based similarity • Sometimes recommendations look like they were random • Sensitive to the parameters • Very few articles are dealing with CF item-to-item recs
  • 27. Case studies on CTR / 1 9/22/2015 CTR almost doubled when we switched from IALS1 to item-kNN on a site where users and items are the same
  • 29. Case studies on CTR / 2 9/22/2015 Comparison of BPR vs. item-kNN on a classified site, for item-to-item recommendations. Item-kNN is the winner
  • 31. Case studies on CTR / 3 9/22/2015 Using BPR vs. item-kNN on a video site for personal recommendations. Measuring number of clicks on recommendations. Result: 4% more clicks for BPR.
  • 33. Case studies on CTR / 4 9/22/2015
  • 34. Conclusions / 1 9/22/2015 Problems with MF: • Lots of parameters to tune • Needs many iteration over the data • if there is no inter-connection between two item sets, they can get similar feature vectors. • Sensitive to noise in data (e.g. users with few events) • Not the best for item-to-item recs, especially when many neighbors exist already
  • 35. Conclusions / 2 9/22/2015 When to use Matrix Factorization • you have one, dense domain (e.g. movies), with not too many items (e.g. less than 100k) • feedback is taste-based • For personalized recommendations (e.g. newsletter) • try both with A/B testing • try blending  Must be smart in blending (e.g. using it for high supported items) • Usually better for offline evaluation metrics