SlideShare a Scribd company logo
1 of 20
Download to read offline
1 KYOTO UNIVERSITY
KYOTO UNIVERSITY
Towards Principled User-side Recommender Systems
Ryoma Sato
2 KYOTO UNIVERSITY
What can we do when the system is unsatisfactory
Solution?
The Recommender system on Twitter is unsatisfactory
Then, stop using Twitter
I want to use Twitter
Then, request a new recommender system to Twitter
It’s hopeless, and it takes long time even if it succeeds
3 KYOTO UNIVERSITY
What can we do when the system is unsatisfactory
Solution
The Recommender system on Twitter is unsatisfactory
Then, stop using Twitter
I want to use Twitter
Then, request a new recommender system to Twitter
It’s hopeless, and it takes long time even if it succeeds
Solution: user-side recommender systems
4 KYOTO UNIVERSITY
User-side RecSys is realized by users

User-side RecSys is a RecSys that a user of
the service builds/realizes

It is in contrast to standard RecSys,
which are realized by the employed engineers

It’s completely different from steerable RecSys

Steerable RecSys are implemented by the employed
engineers, so users need to wait until implemented
 Steerable Recsys enables
users to select interests
Active Solution
Passive Solution
5 KYOTO UNIVERSITY
Challenge: users cannot access to the database

Standard RecSyS
are implemented by Engineers of
Twitter, who have complete
access to the database.

User-side RecSys
are implemented by users, who don’t have direct access to the database.
Twitter Inc.
Database
Twitter Engineer
RecSys
SQL
ANN
Twitter Inc.
Database
User
SQL
Search Query
API
6 KYOTO UNIVERSITY
We consider item-to-item recommendation

Problem setting:
item-to-item recommendation
Movie RecSys @IMDb
Item: movie
Users who watched Toy
Story also watched them.
User RecSys @Twitter
Item: user
Users who follow
@SIAMDataMining also
follow them.
Product RecSys @Amazon
Item: product
Customers who bought this
item also bought
https://www.imdb.com/title/tt0114709/ https://twitter.com/SIAMDataMining https://www.amazon.com/dp/0262035618
7 KYOTO UNIVERSITY
Are web pages enough for building RecSys?

Users observe little information for bulding RecSys
Users observe web pages only
(and estimate information indirectly)
Official developers observe and manipulate DB
Thus they can easily build a RecSys
RQ1: Are these information enough for
users to build RecSys?
8 KYOTO UNIVERSITY
Users can reverse engineer item representations

RQ1: Are these information enough for users to build RecSys?

A: Yes!!
I showed that users can ``reverse-engineer’’ the
representations of items, which are sufficient for building
Recsys.

Idea: The recommendation graph is observable.
This is the k-NN graph of item representations.
The representation can be estimated from kNN graphs
via manifold learning
Official recommendations
build
Recommendation graph
9 KYOTO UNIVERSITY
Use-side Recsys is feasible, but expensive

The result indicates that building a user-side recommender
system is a feasible problem from information-theoretic limit.

Naive user-side recommender algorithm:
1. Visit all item pages
2. Build the recommendation graph
3. Estimate the item representations
4. Build and customize a recommender system based
on the representations
(e.g., by filtering categories and lengthening the slot)

Problem: Too much cost

RQ2: Can we sidestep the direct reverse engineering
and build RecSys more efficiently?
10 KYOTO UNIVERSITY
We want each sensitive attribute to appear equally

Assume that each item has a discrete sensitive attribute and that
the sensitive attributes of the items are observable.

We want each sensitive attribute to appear equally in the results.

What kind of sensitive attribute to use is at the discretion of us.
Movie:
Sensitive attribute = country
User: Sensitive attribute = gender
https://twitter.com/Suzu_Mg https://twitter.com/sudaofficial
https://www.imdb.com/title/tt0114709/
https://www.imdb.com/title/tt0245429/
Another example:
Sensitive Attribute = Popularity ∈ {High, Mid, Low}
(Based on the number of followers)
woman
woman
woman woman
man
man
11 KYOTO UNIVERSITY
We want a user-side RecSys with 3 good props

I propose three desirable properties:

Consistency: If we do not impose fairness, nDCG of a
user-side RecSys should be the same as the official one.

Soundness: If we specify τ, a user-side RecSys should
show at least τ items from each sensitive attribute.

Locality: A user-side RecSys should not download all
pages.

The existing methods and the reverse engineering
approach lack at least one property.

RQ2’: Is there a user-side RecSys that satisfies all?
12 KYOTO UNIVERSITY
I propose a user-side RecSys with three props

RQ2’: Is there a user-side RecSys that satisfies all?

A: Yes!!
I propose Consul, a user-side RecSys that satisfies
all three properties.

Idea: Consul also uses the recommendation graph.
But it traverses the graph locally via DFS.
The official system is
biased to American movies.
I want to
recommend movies
for Toy Story with countries in mind.
13 KYOTO UNIVERSITY
Experiments
14 KYOTO UNIVERSITY
Consul’s recommendations are diverse

Case Study:

I set release dates as the sensitive attribute and built a
user-side movie recommender system with Consul.

The official one recommends contemporary popular
movies.
Consul recommends old sci-fi movies as well (diverse).
The recommendations for
Terminator 2.
15 KYOTO UNIVERSITY
Consul performs well

Quantitative Experiments:

I evaluated existing methods and Consul with standard
benchmarks

The proposed method performs as well as existing
methods
(proposed) →
16 KYOTO UNIVERSITY
Consul is much efficient

Quantitative Experiments:

I evaluated existing methods and Consul with standard
benchmarks

The proposed method is much more efficient than
existing methods thanks to locality
(proposed) →
17 KYOTO UNIVERSITY
Consul is much efficient

User study:

I asked which of the official recommender systems and
the user-side system build by Consul provide more
relevant and diverse recommendations.

About the same number of users chose the official one
and Consul in terms of relevance.
Many users thought Consul was more diverse.
Which system is more relevant?
Which system is more diverse?
18 KYOTO UNIVERSITY
Consul is much efficient

Case study in the wild:

I conducted a case study on real-world Twitter.

I built use-side user recommendation systems.

I’m not an employer of Twitter but an ordinary user.
Nevertheless, I succeeded in building a new RecSys on
Twitter.
Results for Hugh Jackman
I set gender as the sensitive attribute
Consul recommends three men and women
Consul is much efficient (real-time inference)
19 KYOTO UNIVERSITY
Conclusion
20 KYOTO UNIVERSITY
Conclusion

I answered two RQ for user-side RecSys.

RQ1: Are available information enough for users to build
RecSys?
→ Yes, by ``reverse engineering’’ representations.

RQ2: Is there a user-side RecSys that satisfies all the
three properties?
→ Yes. I propose Consul.

The proposed method performs well and is much more
efficient than existing methods.
arXiv: https://arxiv.org/abs/2208.09864 GitHub: https://github.com/joisino/consul

More Related Content

Similar to Towards Principled User-side Recommender Systems

An Introduction to Usability
An Introduction to UsabilityAn Introduction to Usability
An Introduction to Usabilitydirk.swart
 
User Zoom Webinar Monster Aug09 Vf
User Zoom Webinar Monster Aug09 VfUser Zoom Webinar Monster Aug09 Vf
User Zoom Webinar Monster Aug09 VfUserZoom
 
YouTube Title Prediction Using Sentiment Analysis
YouTube Title Prediction Using Sentiment AnalysisYouTube Title Prediction Using Sentiment Analysis
YouTube Title Prediction Using Sentiment AnalysisIRJET Journal
 
MOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERING
MOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERINGMOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERING
MOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERINGIRJET Journal
 
National e-Learning Laboratory
National e-Learning LaboratoryNational e-Learning Laboratory
National e-Learning LaboratoryStephan Weibelzahl
 
Collaborative Filtering Recommendation System
Collaborative Filtering Recommendation SystemCollaborative Filtering Recommendation System
Collaborative Filtering Recommendation SystemMilind Gokhale
 
IRJET- Review on Different Recommendation Techniques for GRS in Online Social...
IRJET- Review on Different Recommendation Techniques for GRS in Online Social...IRJET- Review on Different Recommendation Techniques for GRS in Online Social...
IRJET- Review on Different Recommendation Techniques for GRS in Online Social...IRJET Journal
 
kdd2015-feed (1)
kdd2015-feed (1)kdd2015-feed (1)
kdd2015-feed (1)Guy Lebanon
 
Fuzzy Logic Based Recommender System
Fuzzy Logic Based Recommender SystemFuzzy Logic Based Recommender System
Fuzzy Logic Based Recommender SystemRSIS International
 
Uz big design talk may10
Uz big design talk may10Uz big design talk may10
Uz big design talk may10UserZoom
 
Metrics in usability testing and user experiences
Metrics in usability testing and user experiencesMetrics in usability testing and user experiences
Metrics in usability testing and user experiencesHim Chitchat
 
Generative AI_ The force-multiplier for SDLC.pptx
Generative AI_ The force-multiplier for SDLC.pptxGenerative AI_ The force-multiplier for SDLC.pptx
Generative AI_ The force-multiplier for SDLC.pptxKumar Iyer
 
Intro to Data Analytics with Oscar's Director of Product
 Intro to Data Analytics with Oscar's Director of Product Intro to Data Analytics with Oscar's Director of Product
Intro to Data Analytics with Oscar's Director of ProductProduct School
 
Building and Scaling High Performing Technology Organizations by Jez Humble a...
Building and Scaling High Performing Technology Organizations by Jez Humble a...Building and Scaling High Performing Technology Organizations by Jez Humble a...
Building and Scaling High Performing Technology Organizations by Jez Humble a...Agile India
 
Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...
Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...
Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...nealw6971
 
Study of Recommendation System Used In Tourism and Travel
Study of Recommendation System Used In Tourism and TravelStudy of Recommendation System Used In Tourism and Travel
Study of Recommendation System Used In Tourism and Travelijtsrd
 
Recommender system and big data (design a smartphone recommender system based...
Recommender system and big data (design a smartphone recommender system based...Recommender system and big data (design a smartphone recommender system based...
Recommender system and big data (design a smartphone recommender system based...Siwar Abidi
 

Similar to Towards Principled User-side Recommender Systems (20)

An Introduction to Usability
An Introduction to UsabilityAn Introduction to Usability
An Introduction to Usability
 
User Zoom Webinar Monster Aug09 Vf
User Zoom Webinar Monster Aug09 VfUser Zoom Webinar Monster Aug09 Vf
User Zoom Webinar Monster Aug09 Vf
 
YouTube Title Prediction Using Sentiment Analysis
YouTube Title Prediction Using Sentiment AnalysisYouTube Title Prediction Using Sentiment Analysis
YouTube Title Prediction Using Sentiment Analysis
 
MOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERING
MOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERINGMOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERING
MOVIE RECOMMENDATION SYSTEM USING COLLABORATIVE FILTERING
 
Test automation Anecdotes
Test automation AnecdotesTest automation Anecdotes
Test automation Anecdotes
 
National e-Learning Laboratory
National e-Learning LaboratoryNational e-Learning Laboratory
National e-Learning Laboratory
 
The UX Analyst
The UX AnalystThe UX Analyst
The UX Analyst
 
Collaborative Filtering Recommendation System
Collaborative Filtering Recommendation SystemCollaborative Filtering Recommendation System
Collaborative Filtering Recommendation System
 
IRJET- Review on Different Recommendation Techniques for GRS in Online Social...
IRJET- Review on Different Recommendation Techniques for GRS in Online Social...IRJET- Review on Different Recommendation Techniques for GRS in Online Social...
IRJET- Review on Different Recommendation Techniques for GRS in Online Social...
 
kdd2015-feed (1)
kdd2015-feed (1)kdd2015-feed (1)
kdd2015-feed (1)
 
Fuzzy Logic Based Recommender System
Fuzzy Logic Based Recommender SystemFuzzy Logic Based Recommender System
Fuzzy Logic Based Recommender System
 
20120140506003
2012014050600320120140506003
20120140506003
 
Uz big design talk may10
Uz big design talk may10Uz big design talk may10
Uz big design talk may10
 
Metrics in usability testing and user experiences
Metrics in usability testing and user experiencesMetrics in usability testing and user experiences
Metrics in usability testing and user experiences
 
Generative AI_ The force-multiplier for SDLC.pptx
Generative AI_ The force-multiplier for SDLC.pptxGenerative AI_ The force-multiplier for SDLC.pptx
Generative AI_ The force-multiplier for SDLC.pptx
 
Intro to Data Analytics with Oscar's Director of Product
 Intro to Data Analytics with Oscar's Director of Product Intro to Data Analytics with Oscar's Director of Product
Intro to Data Analytics with Oscar's Director of Product
 
Building and Scaling High Performing Technology Organizations by Jez Humble a...
Building and Scaling High Performing Technology Organizations by Jez Humble a...Building and Scaling High Performing Technology Organizations by Jez Humble a...
Building and Scaling High Performing Technology Organizations by Jez Humble a...
 
Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...
Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...
Designed by Committee: An Analytics and User-Focused Approach to the Overhaul...
 
Study of Recommendation System Used In Tourism and Travel
Study of Recommendation System Used In Tourism and TravelStudy of Recommendation System Used In Tourism and Travel
Study of Recommendation System Used In Tourism and Travel
 
Recommender system and big data (design a smartphone recommender system based...
Recommender system and big data (design a smartphone recommender system based...Recommender system and big data (design a smartphone recommender system based...
Recommender system and big data (design a smartphone recommender system based...
 

More from joisino

キャッシュオブリビアスアルゴリズム
キャッシュオブリビアスアルゴリズムキャッシュオブリビアスアルゴリズム
キャッシュオブリビアスアルゴリズムjoisino
 
Metric Recovery from Unweighted k-NN Graphs
Metric Recovery from Unweighted k-NN GraphsMetric Recovery from Unweighted k-NN Graphs
Metric Recovery from Unweighted k-NN Graphsjoisino
 
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...joisino
 
An Introduction to Spectral Graph Theory
An Introduction to Spectral Graph TheoryAn Introduction to Spectral Graph Theory
An Introduction to Spectral Graph Theoryjoisino
 
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...joisino
 
最適輸送入門
最適輸送入門最適輸送入門
最適輸送入門joisino
 
ユーザーサイド情報検索システム
ユーザーサイド情報検索システムユーザーサイド情報検索システム
ユーザーサイド情報検索システムjoisino
 
最適輸送の解き方
最適輸送の解き方最適輸送の解き方
最適輸送の解き方joisino
 
Random Features Strengthen Graph Neural Networks
Random Features Strengthen Graph Neural NetworksRandom Features Strengthen Graph Neural Networks
Random Features Strengthen Graph Neural Networksjoisino
 
Fast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a TreeFast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a Treejoisino
 
グラフニューラルネットワークとグラフ組合せ問題
グラフニューラルネットワークとグラフ組合せ問題グラフニューラルネットワークとグラフ組合せ問題
グラフニューラルネットワークとグラフ組合せ問題joisino
 
死にたくない
死にたくない死にたくない
死にたくないjoisino
 

More from joisino (12)

キャッシュオブリビアスアルゴリズム
キャッシュオブリビアスアルゴリズムキャッシュオブリビアスアルゴリズム
キャッシュオブリビアスアルゴリズム
 
Metric Recovery from Unweighted k-NN Graphs
Metric Recovery from Unweighted k-NN GraphsMetric Recovery from Unweighted k-NN Graphs
Metric Recovery from Unweighted k-NN Graphs
 
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
 
An Introduction to Spectral Graph Theory
An Introduction to Spectral Graph TheoryAn Introduction to Spectral Graph Theory
An Introduction to Spectral Graph Theory
 
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...
 
最適輸送入門
最適輸送入門最適輸送入門
最適輸送入門
 
ユーザーサイド情報検索システム
ユーザーサイド情報検索システムユーザーサイド情報検索システム
ユーザーサイド情報検索システム
 
最適輸送の解き方
最適輸送の解き方最適輸送の解き方
最適輸送の解き方
 
Random Features Strengthen Graph Neural Networks
Random Features Strengthen Graph Neural NetworksRandom Features Strengthen Graph Neural Networks
Random Features Strengthen Graph Neural Networks
 
Fast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a TreeFast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a Tree
 
グラフニューラルネットワークとグラフ組合せ問題
グラフニューラルネットワークとグラフ組合せ問題グラフニューラルネットワークとグラフ組合せ問題
グラフニューラルネットワークとグラフ組合せ問題
 
死にたくない
死にたくない死にたくない
死にたくない
 

Recently uploaded

Artificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C PArtificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C PPRINCE C P
 
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...jana861314
 
Bentham & Hooker's Classification. along with the merits and demerits of the ...
Bentham & Hooker's Classification. along with the merits and demerits of the ...Bentham & Hooker's Classification. along with the merits and demerits of the ...
Bentham & Hooker's Classification. along with the merits and demerits of the ...Nistarini College, Purulia (W.B) India
 
A relative description on Sonoporation.pdf
A relative description on Sonoporation.pdfA relative description on Sonoporation.pdf
A relative description on Sonoporation.pdfnehabiju2046
 
Work, Energy and Power for class 10 ICSE Physics
Work, Energy and Power for class 10 ICSE PhysicsWork, Energy and Power for class 10 ICSE Physics
Work, Energy and Power for class 10 ICSE Physicsvishikhakeshava1
 
Physiochemical properties of nanomaterials and its nanotoxicity.pptx
Physiochemical properties of nanomaterials and its nanotoxicity.pptxPhysiochemical properties of nanomaterials and its nanotoxicity.pptx
Physiochemical properties of nanomaterials and its nanotoxicity.pptxAArockiyaNisha
 
G9 Science Q4- Week 1-2 Projectile Motion.ppt
G9 Science Q4- Week 1-2 Projectile Motion.pptG9 Science Q4- Week 1-2 Projectile Motion.ppt
G9 Science Q4- Week 1-2 Projectile Motion.pptMAESTRELLAMesa2
 
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptxUnlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptxanandsmhk
 
Nanoparticles synthesis and characterization​ ​
Nanoparticles synthesis and characterization​  ​Nanoparticles synthesis and characterization​  ​
Nanoparticles synthesis and characterization​ ​kaibalyasahoo82800
 
Boyles law module in the grade 10 science
Boyles law module in the grade 10 scienceBoyles law module in the grade 10 science
Boyles law module in the grade 10 sciencefloriejanemacaya1
 
Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?Patrick Diehl
 
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCRStunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCRDelhi Call girls
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Lokesh Kothari
 
Luciferase in rDNA technology (biotechnology).pptx
Luciferase in rDNA technology (biotechnology).pptxLuciferase in rDNA technology (biotechnology).pptx
Luciferase in rDNA technology (biotechnology).pptxAleenaTreesaSaji
 
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...anilsa9823
 
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxSOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxkessiyaTpeter
 
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bNightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bSérgio Sacani
 
GFP in rDNA Technology (Biotechnology).pptx
GFP in rDNA Technology (Biotechnology).pptxGFP in rDNA Technology (Biotechnology).pptx
GFP in rDNA Technology (Biotechnology).pptxAleenaTreesaSaji
 
Animal Communication- Auditory and Visual.pptx
Animal Communication- Auditory and Visual.pptxAnimal Communication- Auditory and Visual.pptx
Animal Communication- Auditory and Visual.pptxUmerFayaz5
 

Recently uploaded (20)

Artificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C PArtificial Intelligence In Microbiology by Dr. Prince C P
Artificial Intelligence In Microbiology by Dr. Prince C P
 
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
 
Bentham & Hooker's Classification. along with the merits and demerits of the ...
Bentham & Hooker's Classification. along with the merits and demerits of the ...Bentham & Hooker's Classification. along with the merits and demerits of the ...
Bentham & Hooker's Classification. along with the merits and demerits of the ...
 
A relative description on Sonoporation.pdf
A relative description on Sonoporation.pdfA relative description on Sonoporation.pdf
A relative description on Sonoporation.pdf
 
Work, Energy and Power for class 10 ICSE Physics
Work, Energy and Power for class 10 ICSE PhysicsWork, Energy and Power for class 10 ICSE Physics
Work, Energy and Power for class 10 ICSE Physics
 
Physiochemical properties of nanomaterials and its nanotoxicity.pptx
Physiochemical properties of nanomaterials and its nanotoxicity.pptxPhysiochemical properties of nanomaterials and its nanotoxicity.pptx
Physiochemical properties of nanomaterials and its nanotoxicity.pptx
 
G9 Science Q4- Week 1-2 Projectile Motion.ppt
G9 Science Q4- Week 1-2 Projectile Motion.pptG9 Science Q4- Week 1-2 Projectile Motion.ppt
G9 Science Q4- Week 1-2 Projectile Motion.ppt
 
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptxUnlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
 
Nanoparticles synthesis and characterization​ ​
Nanoparticles synthesis and characterization​  ​Nanoparticles synthesis and characterization​  ​
Nanoparticles synthesis and characterization​ ​
 
Boyles law module in the grade 10 science
Boyles law module in the grade 10 scienceBoyles law module in the grade 10 science
Boyles law module in the grade 10 science
 
Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?Is RISC-V ready for HPC workload? Maybe?
Is RISC-V ready for HPC workload? Maybe?
 
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCRStunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
 
Luciferase in rDNA technology (biotechnology).pptx
Luciferase in rDNA technology (biotechnology).pptxLuciferase in rDNA technology (biotechnology).pptx
Luciferase in rDNA technology (biotechnology).pptx
 
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
 
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxSOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
 
The Philosophy of Science
The Philosophy of ScienceThe Philosophy of Science
The Philosophy of Science
 
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bNightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
 
GFP in rDNA Technology (Biotechnology).pptx
GFP in rDNA Technology (Biotechnology).pptxGFP in rDNA Technology (Biotechnology).pptx
GFP in rDNA Technology (Biotechnology).pptx
 
Animal Communication- Auditory and Visual.pptx
Animal Communication- Auditory and Visual.pptxAnimal Communication- Auditory and Visual.pptx
Animal Communication- Auditory and Visual.pptx
 

Towards Principled User-side Recommender Systems

  • 1. 1 KYOTO UNIVERSITY KYOTO UNIVERSITY Towards Principled User-side Recommender Systems Ryoma Sato
  • 2. 2 KYOTO UNIVERSITY What can we do when the system is unsatisfactory Solution? The Recommender system on Twitter is unsatisfactory Then, stop using Twitter I want to use Twitter Then, request a new recommender system to Twitter It’s hopeless, and it takes long time even if it succeeds
  • 3. 3 KYOTO UNIVERSITY What can we do when the system is unsatisfactory Solution The Recommender system on Twitter is unsatisfactory Then, stop using Twitter I want to use Twitter Then, request a new recommender system to Twitter It’s hopeless, and it takes long time even if it succeeds Solution: user-side recommender systems
  • 4. 4 KYOTO UNIVERSITY User-side RecSys is realized by users  User-side RecSys is a RecSys that a user of the service builds/realizes  It is in contrast to standard RecSys, which are realized by the employed engineers  It’s completely different from steerable RecSys  Steerable RecSys are implemented by the employed engineers, so users need to wait until implemented  Steerable Recsys enables users to select interests Active Solution Passive Solution
  • 5. 5 KYOTO UNIVERSITY Challenge: users cannot access to the database  Standard RecSyS are implemented by Engineers of Twitter, who have complete access to the database.  User-side RecSys are implemented by users, who don’t have direct access to the database. Twitter Inc. Database Twitter Engineer RecSys SQL ANN Twitter Inc. Database User SQL Search Query API
  • 6. 6 KYOTO UNIVERSITY We consider item-to-item recommendation  Problem setting: item-to-item recommendation Movie RecSys @IMDb Item: movie Users who watched Toy Story also watched them. User RecSys @Twitter Item: user Users who follow @SIAMDataMining also follow them. Product RecSys @Amazon Item: product Customers who bought this item also bought https://www.imdb.com/title/tt0114709/ https://twitter.com/SIAMDataMining https://www.amazon.com/dp/0262035618
  • 7. 7 KYOTO UNIVERSITY Are web pages enough for building RecSys?  Users observe little information for bulding RecSys Users observe web pages only (and estimate information indirectly) Official developers observe and manipulate DB Thus they can easily build a RecSys RQ1: Are these information enough for users to build RecSys?
  • 8. 8 KYOTO UNIVERSITY Users can reverse engineer item representations  RQ1: Are these information enough for users to build RecSys?  A: Yes!! I showed that users can ``reverse-engineer’’ the representations of items, which are sufficient for building Recsys.  Idea: The recommendation graph is observable. This is the k-NN graph of item representations. The representation can be estimated from kNN graphs via manifold learning Official recommendations build Recommendation graph
  • 9. 9 KYOTO UNIVERSITY Use-side Recsys is feasible, but expensive  The result indicates that building a user-side recommender system is a feasible problem from information-theoretic limit.  Naive user-side recommender algorithm: 1. Visit all item pages 2. Build the recommendation graph 3. Estimate the item representations 4. Build and customize a recommender system based on the representations (e.g., by filtering categories and lengthening the slot)  Problem: Too much cost  RQ2: Can we sidestep the direct reverse engineering and build RecSys more efficiently?
  • 10. 10 KYOTO UNIVERSITY We want each sensitive attribute to appear equally  Assume that each item has a discrete sensitive attribute and that the sensitive attributes of the items are observable.  We want each sensitive attribute to appear equally in the results.  What kind of sensitive attribute to use is at the discretion of us. Movie: Sensitive attribute = country User: Sensitive attribute = gender https://twitter.com/Suzu_Mg https://twitter.com/sudaofficial https://www.imdb.com/title/tt0114709/ https://www.imdb.com/title/tt0245429/ Another example: Sensitive Attribute = Popularity ∈ {High, Mid, Low} (Based on the number of followers) woman woman woman woman man man
  • 11. 11 KYOTO UNIVERSITY We want a user-side RecSys with 3 good props  I propose three desirable properties:  Consistency: If we do not impose fairness, nDCG of a user-side RecSys should be the same as the official one.  Soundness: If we specify τ, a user-side RecSys should show at least τ items from each sensitive attribute.  Locality: A user-side RecSys should not download all pages.  The existing methods and the reverse engineering approach lack at least one property.  RQ2’: Is there a user-side RecSys that satisfies all?
  • 12. 12 KYOTO UNIVERSITY I propose a user-side RecSys with three props  RQ2’: Is there a user-side RecSys that satisfies all?  A: Yes!! I propose Consul, a user-side RecSys that satisfies all three properties.  Idea: Consul also uses the recommendation graph. But it traverses the graph locally via DFS. The official system is biased to American movies. I want to recommend movies for Toy Story with countries in mind.
  • 14. 14 KYOTO UNIVERSITY Consul’s recommendations are diverse  Case Study:  I set release dates as the sensitive attribute and built a user-side movie recommender system with Consul.  The official one recommends contemporary popular movies. Consul recommends old sci-fi movies as well (diverse). The recommendations for Terminator 2.
  • 15. 15 KYOTO UNIVERSITY Consul performs well  Quantitative Experiments:  I evaluated existing methods and Consul with standard benchmarks  The proposed method performs as well as existing methods (proposed) →
  • 16. 16 KYOTO UNIVERSITY Consul is much efficient  Quantitative Experiments:  I evaluated existing methods and Consul with standard benchmarks  The proposed method is much more efficient than existing methods thanks to locality (proposed) →
  • 17. 17 KYOTO UNIVERSITY Consul is much efficient  User study:  I asked which of the official recommender systems and the user-side system build by Consul provide more relevant and diverse recommendations.  About the same number of users chose the official one and Consul in terms of relevance. Many users thought Consul was more diverse. Which system is more relevant? Which system is more diverse?
  • 18. 18 KYOTO UNIVERSITY Consul is much efficient  Case study in the wild:  I conducted a case study on real-world Twitter.  I built use-side user recommendation systems.  I’m not an employer of Twitter but an ordinary user. Nevertheless, I succeeded in building a new RecSys on Twitter. Results for Hugh Jackman I set gender as the sensitive attribute Consul recommends three men and women Consul is much efficient (real-time inference)
  • 20. 20 KYOTO UNIVERSITY Conclusion  I answered two RQ for user-side RecSys.  RQ1: Are available information enough for users to build RecSys? → Yes, by ``reverse engineering’’ representations.  RQ2: Is there a user-side RecSys that satisfies all the three properties? → Yes. I propose Consul.  The proposed method performs well and is much more efficient than existing methods. arXiv: https://arxiv.org/abs/2208.09864 GitHub: https://github.com/joisino/consul