SlideShare ist ein Scribd-Unternehmen logo
1 von 44
Downloaden Sie, um offline zu lesen
A Human Perspective on
Algorithmic Similarity
RecSys 2020
Zach Schendel, Faraz Farzin, & Siddhi Sundar
Netflix Product Innovation, Consumer Insights
Even though it was highly recommended by a Netflix
algorithm, I rejected this movie many times in 2017.
It just didn’t look or sound
good from the description.
Importantly, I like to click on “More Like This” because
I want to find out what unfamiliar movies are similar to.
I have seen a few of these.
They were decent, but the
connection between them
was not obvious
I eventually watched it after reading more about it
online. It remains my favorite Netflix Original movie!
...but, if I were to personalize
“More Like This”, I would include
similar titles that would be
explanatory and authentic for me
in that moment.
And… I wasn’t the only one who had an opinion about
similar recommendations on Netflix...
These and other examples got us thinking… what
exactly is similarity? How similar does something
need to be to describe another movie? When do
algorithms need to be restrictive, and when do they
have permission to be relaxed?
Where is the line between “similar” and “dissimilar”?
We decided to ask Netflix members!
We employed three different research methods.
An international landscape
assessment of how similarity is
used in and outside of the Netflix
category by other businesses.
Qualitative interviews with Netflix
members assessing how
appropriate and successful our
recommendations are in different
places where similarity could be
used as a driver for algorithms.
A quantitative evaluation of
perceived similarity between a series
of source titles and potential
recommendations using a technique
called inverse Multi-dimensional
Scaling (iMDS) [1].
More on this later...
[1] Kriegeskorte & Mur, 2012
1. 2. 3.
the Person
Who is seeing the recommendations, and what
is their past experience with the source?
the Context
What is going on at the moment? What are
the user’s needs?
1 2 3the Placement
Where are the recommendations placed within
the Netflix user interface?
The answer...it’s complicated! But we can really help our
recommendations! There are 3 sources of complexity:
the Person the Context
1 2 3the Placement
Where are the recommendations placed within
the Netflix user interface?
Many places in the Netflix interface are populated by
algorithms that factor in some dimension of similarity.
We found that members had higher expectations of
similarity when part of a 1:1 recommendation.
but… this doesn’t make much sense
at all. The Crown and The Office have
little in common.
This makes perfect sense. The Kissing
Booth and To All the Boys... are teen
romantic dramas with female leads.
Both cases are risky - there are no backups/other
titles to help explain the similarity.
But, there were lower expectations in places with
1:many recommendations, like while browsing.
Members are unlikely to say, “oh you liked Million Dollar Beach House? You should definitely watch Queer Eye.”
...but this broad placement is mostly made up of reality shows, so the link makes complete sense as a row.
And, like earlier, there were higher expectations in
placements that explicitly result from member action.
If they search for something specific
If they click into a specific title
and navigate to “More Like This”
Members are looking for something specifically similar.
the Person the Context
1 2 3the Placement
There is no one-size-fits all approach to employing
similarity signals in algorithms in all candidate
placements. Places that display 1:1
recommendations or react to explicit user
engagement have higher expectations of similarity.
Summary:
the Placement the Context
1 2 3the Person
Who is seeing the recommendations, and what
is their past experience with the source?
Method: We asked members to use iMDS[1] to
self-cluster content by similarity.
[1] Kriegeskorte & Mur, 2012
More similar =
closer together.
Will be added
Let’s walk through one example to illustrate what we
learned. Stranger Things is multidimensional.
Sci-fi . Fantasy . Teen . Horror . Thriller . 80’s . Nostalgia . Starring
Winona Ryder . Ominous . Scary . Exciting . Comedy . Coming of age
These movies are very similar to Stranger Things on
almost every one of those dimensions.
These high-similarity options would work well in 1:1 placements discussed above.
But, you quickly run out of options, and these won’t make up the bulk of placements.
Both broad and specific drivers of perceived similarity
can be used to fill out the bulk of placements.
Sci-fi . Fantasy . Teen . Horror . Thriller . 80’s . Nostalgia . Starring
Winona Ryder . Ominous . Scary . Exciting . Comedy . Coming of age
Broad similarity drivers were surface level, like genre.
They piqued interest and had clear source links.
Sci-fi . Fantasy . Teen . Horror . Thriller . 80’s . Nostalgia . Starring
Winona Ryder . Ominous . Scary . Exciting . Comedy . Coming of age
Specific similarity drivers were varied and difficult to
predict but were more salient final points of proof.
it stars Winona
Ryder!
‘80s nostalgia! Battling science
monsters!
Group of misfit
teens in the 80’s!
But...they are high-risk/high-reward. Trustbusters
often occur when a wrong or unclear link are chosen.
it stars Winona
Ryder!
‘80s nostalgia! Battling science
monsters!
Group of misfit
teens in the 80’s!
the Person
The degree to which something is or is not similar
is in the eyes of the beholder, all of whom might
latch on to different paths or give different
permissions. The best path for algorithms is to find
ways to balance broad and specific drivers of
similarity. Broadly similar recommendations that
differentially emphasize specific drivers for an
individual should minimize trustbusters.
the Context
1 2 3the Placement
There is no one-size-fits all approach to employing
similarity signals in algorithms in all candidate
placements. Places that display 1:1
recommendations or react to explicit user
engagement have higher expectations of similarity.
Summary:
the Placement the Person
1 2 3the Context
What is going on at the moment? What are
the user’s needs?
Finally, to add even more complexity, even if you hold
placement and person consistent, context also matters.
After you finish a movie or show,
Netflix recommends something.
You finished... ...try this next.
After finishing a show, members are most likely to
watch something similar.
18%...watch another
reality show
#2 among options
But while this is certainly a common path, it is not
the most common or only path.
52%...watch another
Netflix Original
#1 among options
One example… After finishing a Netflix Original reality TV show...
55%...watch more
unserialized content
#1 among options
There are multiple contexts in which similarity is
unnecessary and in fact, the opposite of ideal.
“That was intense. I need
a change of pace,
something lighter”
“I don’t have time for
another movie. I’ll just put
on something short.”
… and more
“I just need to rewatch
something familiar, something
I have watched before.”
the Placement the Person
1 2 3the Context
Even if the placement and the person are
held constant, context further impacts the
perception or need for similarity. Algos
that can take context into consideration
(e.g., today it’s likely to be more of the
same day, tomorrow a change of pace) or
allow for balance across multiple paths
should be more successful.
Summary:
% Fewer Perceived Trustbusters comparing the old similarity
model to the new similarity model by similarity rank.
FEWERTrustbusters
In the end, the research validated a new similarity
model! There were fewer perceived trustbusters!
And, for me at least, these are better recommendations!
“More Like This” Similars
Thanks!
A Human Perspective on Algorithmic Similarity
RecSys 2020
Zach Schendel, Faraz Farzin, & Siddhi Sundar
Netflix Product Innovation, Consumer Insights
Slides for 3 minute intro video
A Human Perspective on
Algorithmic Similarity
RecSys 2020
Zach Schendel, Faraz Farzin, & Siddhi Sundar
Netflix Product Innovation, Consumer Insights
If you look at Twitter, there are
typically 3 types of comments.
I really liked this!
… I want more like it!1
uuuuh…
I don’t understand2
and…
a pleasant surprise!3
“More Like This” Similars
We are going to talk about how Netflix worked
directly with member research to get from this?!?...
?!?
...to this!...
“More Like This” Similars!
the Person
Who is seeing the recommendations, and what
is their past experience with the source?
the Context
What is going on at the moment? What are
the user’s needs?
1 2 3the Placement
Where are the recommendations placed within
the Netflix user interface?
...by exploring how perceived similarity can be strongly
influenced by three new variables:
20200903T225500Z__recsys__IN1032__a-human-perspective-on-algorit
SlidesLive ID

Weitere ähnliche Inhalte

Was ist angesagt?

Past, Present & Future of Recommender Systems: An Industry Perspective
Past, Present & Future of Recommender Systems: An Industry PerspectivePast, Present & Future of Recommender Systems: An Industry Perspective
Past, Present & Future of Recommender Systems: An Industry PerspectiveJustin Basilico
 
Recent Trends in Personalization: A Netflix Perspective
Recent Trends in Personalization: A Netflix PerspectiveRecent Trends in Personalization: A Netflix Perspective
Recent Trends in Personalization: A Netflix PerspectiveJustin Basilico
 
Supporting decisions with ML
Supporting decisions with MLSupporting decisions with ML
Supporting decisions with MLMegan Neider
 
Tableau Conference 2018: Binging on Data - Enabling Analytics at Netflix
Tableau Conference 2018: Binging on Data - Enabling Analytics at NetflixTableau Conference 2018: Binging on Data - Enabling Analytics at Netflix
Tableau Conference 2018: Binging on Data - Enabling Analytics at NetflixBlake Irvine
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsJustin Basilico
 
Contextualization at Netflix
Contextualization at NetflixContextualization at Netflix
Contextualization at NetflixLinas Baltrunas
 
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
 Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se... Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...Sudeep Das, Ph.D.
 
Sequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsSequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsJaya Kawale
 
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
 
Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations 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
 
Homepage Personalization at Spotify
Homepage Personalization at SpotifyHomepage Personalization at Spotify
Homepage Personalization at SpotifyOguz Semerci
 
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
 
Calibrated Recommendations
Calibrated RecommendationsCalibrated Recommendations
Calibrated RecommendationsHarald Steck
 
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
 
Making Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableMaking Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableJustin Basilico
 
Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)Xavier Amatriain
 

Was ist angesagt? (20)

Past, Present & Future of Recommender Systems: An Industry Perspective
Past, Present & Future of Recommender Systems: An Industry PerspectivePast, Present & Future of Recommender Systems: An Industry Perspective
Past, Present & Future of Recommender Systems: An Industry Perspective
 
Recent Trends in Personalization: A Netflix Perspective
Recent Trends in Personalization: A Netflix PerspectiveRecent Trends in Personalization: A Netflix Perspective
Recent Trends in Personalization: A Netflix Perspective
 
Supporting decisions with ML
Supporting decisions with MLSupporting decisions with ML
Supporting decisions with ML
 
Tableau Conference 2018: Binging on Data - Enabling Analytics at Netflix
Tableau Conference 2018: Binging on Data - Enabling Analytics at NetflixTableau Conference 2018: Binging on Data - Enabling Analytics at Netflix
Tableau Conference 2018: Binging on Data - Enabling Analytics at Netflix
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Contextualization at Netflix
Contextualization at NetflixContextualization at Netflix
Contextualization at Netflix
 
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
 Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se... Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
Deeper Things: How Netflix Leverages Deep Learning in Recommendations and Se...
 
Sequential Decision Making in Recommendations
Sequential Decision Making in RecommendationsSequential Decision Making in Recommendations
Sequential Decision Making in Recommendations
 
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
 
Context Aware Recommendations at Netflix
Context Aware Recommendations at NetflixContext Aware Recommendations at Netflix
Context Aware Recommendations 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
 
Homepage Personalization at Spotify
Homepage Personalization at SpotifyHomepage Personalization at Spotify
Homepage Personalization at Spotify
 
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
 
Calibrated Recommendations
Calibrated RecommendationsCalibrated Recommendations
Calibrated Recommendations
 
Learning to Personalize
Learning to PersonalizeLearning to Personalize
Learning to Personalize
 
Recommender Systems
Recommender SystemsRecommender Systems
Recommender Systems
 
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
 
Recommender Systems
Recommender SystemsRecommender Systems
Recommender Systems
 
Making Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms ReliableMaking Netflix Machine Learning Algorithms Reliable
Making Netflix Machine Learning Algorithms Reliable
 
Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)
 

Ähnlich wie A Human Perspective on Algorithmic Similarity

Writing for new media
Writing for new mediaWriting for new media
Writing for new mediaDori Adar
 
Top Essay Sites. 11 Best College Application Essay Examples Format Guide
Top Essay Sites. 11 Best College Application Essay Examples Format GuideTop Essay Sites. 11 Best College Application Essay Examples Format Guide
Top Essay Sites. 11 Best College Application Essay Examples Format GuideCaroline Barnett
 
The most appreciated codes and conventions in horror
The most appreciated codes and conventions in horrorThe most appreciated codes and conventions in horror
The most appreciated codes and conventions in horrorlavers1234
 
Quantitative Audience Research
Quantitative Audience ResearchQuantitative Audience Research
Quantitative Audience Researchavapetal
 
Girlfriend Love Essay. Online assignment writing service.
Girlfriend Love Essay. Online assignment writing service.Girlfriend Love Essay. Online assignment writing service.
Girlfriend Love Essay. Online assignment writing service.Diana Hole
 
Question 8 of media questions (2)
Question 8 of media questions (2)Question 8 of media questions (2)
Question 8 of media questions (2)edwardwhiley
 
Question 8 of media questions
Question 8 of media questionsQuestion 8 of media questions
Question 8 of media questionsedwardwhiley
 
Target audience research
Target audience researchTarget audience research
Target audience researchnykelly
 
The Architecture of Social Websites: Reputation
The Architecture of Social Websites: ReputationThe Architecture of Social Websites: Reputation
The Architecture of Social Websites: ReputationBryce Glass
 
Sample Apa Annotated Bibliography Best Of Accur
Sample Apa Annotated Bibliography Best Of AccurSample Apa Annotated Bibliography Best Of Accur
Sample Apa Annotated Bibliography Best Of AccurRuth Uithoven
 
Preliminary Data
Preliminary DataPreliminary Data
Preliminary DataRoxy1m7
 
Policy Brief Template Word Free - Printable Templates
Policy Brief Template Word Free - Printable TemplatesPolicy Brief Template Word Free - Printable Templates
Policy Brief Template Word Free - Printable TemplatesJennifer Wood
 
Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...
Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...
Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...Sri Ambati
 
Audience quantitative research & analysis
Audience quantitative research & analysisAudience quantitative research & analysis
Audience quantitative research & analysisKyri Lambis
 
Audience Quantitative Research & Analysis
Audience Quantitative Research & AnalysisAudience Quantitative Research & Analysis
Audience Quantitative Research & AnalysisKyri Lambis
 
Essay On Anna Hazare In Hindi. Online assignment writing service.
Essay On Anna Hazare In Hindi. Online assignment writing service.Essay On Anna Hazare In Hindi. Online assignment writing service.
Essay On Anna Hazare In Hindi. Online assignment writing service.Beth Garcia
 
Crazy Futures: Why Plausibility is Maladaptive
Crazy Futures: Why Plausibility is MaladaptiveCrazy Futures: Why Plausibility is Maladaptive
Crazy Futures: Why Plausibility is MaladaptiveWendy Schultz
 
Essay On Mercy Killing.pdf
Essay On Mercy Killing.pdfEssay On Mercy Killing.pdf
Essay On Mercy Killing.pdfLory Holets
 
Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16
Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16
Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16Kimberly Gomez
 

Ähnlich wie A Human Perspective on Algorithmic Similarity (20)

Writing for new media
Writing for new mediaWriting for new media
Writing for new media
 
Top Essay Sites. 11 Best College Application Essay Examples Format Guide
Top Essay Sites. 11 Best College Application Essay Examples Format GuideTop Essay Sites. 11 Best College Application Essay Examples Format Guide
Top Essay Sites. 11 Best College Application Essay Examples Format Guide
 
The most appreciated codes and conventions in horror
The most appreciated codes and conventions in horrorThe most appreciated codes and conventions in horror
The most appreciated codes and conventions in horror
 
Quantitative Audience Research
Quantitative Audience ResearchQuantitative Audience Research
Quantitative Audience Research
 
Girlfriend Love Essay. Online assignment writing service.
Girlfriend Love Essay. Online assignment writing service.Girlfriend Love Essay. Online assignment writing service.
Girlfriend Love Essay. Online assignment writing service.
 
Question 8 of media questions (2)
Question 8 of media questions (2)Question 8 of media questions (2)
Question 8 of media questions (2)
 
Question 8 of media questions
Question 8 of media questionsQuestion 8 of media questions
Question 8 of media questions
 
Target audience research
Target audience researchTarget audience research
Target audience research
 
The Architecture of Social Websites: Reputation
The Architecture of Social Websites: ReputationThe Architecture of Social Websites: Reputation
The Architecture of Social Websites: Reputation
 
Sample Apa Annotated Bibliography Best Of Accur
Sample Apa Annotated Bibliography Best Of AccurSample Apa Annotated Bibliography Best Of Accur
Sample Apa Annotated Bibliography Best Of Accur
 
Preliminary Data
Preliminary DataPreliminary Data
Preliminary Data
 
Policy Brief Template Word Free - Printable Templates
Policy Brief Template Word Free - Printable TemplatesPolicy Brief Template Word Free - Printable Templates
Policy Brief Template Word Free - Printable Templates
 
Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...
Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...
Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant P...
 
Audience quantitative research & analysis
Audience quantitative research & analysisAudience quantitative research & analysis
Audience quantitative research & analysis
 
Audience Quantitative Research & Analysis
Audience Quantitative Research & AnalysisAudience Quantitative Research & Analysis
Audience Quantitative Research & Analysis
 
Essay On Anna Hazare In Hindi. Online assignment writing service.
Essay On Anna Hazare In Hindi. Online assignment writing service.Essay On Anna Hazare In Hindi. Online assignment writing service.
Essay On Anna Hazare In Hindi. Online assignment writing service.
 
Data presentation
Data presentationData presentation
Data presentation
 
Crazy Futures: Why Plausibility is Maladaptive
Crazy Futures: Why Plausibility is MaladaptiveCrazy Futures: Why Plausibility is Maladaptive
Crazy Futures: Why Plausibility is Maladaptive
 
Essay On Mercy Killing.pdf
Essay On Mercy Killing.pdfEssay On Mercy Killing.pdf
Essay On Mercy Killing.pdf
 
Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16
Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16
Size Of Writing Paper. Writing Paper Sizes Chart. 2019-01-16
 

Kürzlich hochgeladen

Predicting Salary Using Data Science: A Comprehensive Analysis.pdf
Predicting Salary Using Data Science: A Comprehensive Analysis.pdfPredicting Salary Using Data Science: A Comprehensive Analysis.pdf
Predicting Salary Using Data Science: A Comprehensive Analysis.pdfBoston Institute of Analytics
 
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...limedy534
 
Generative AI for Social Good at Open Data Science East 2024
Generative AI for Social Good at Open Data Science East 2024Generative AI for Social Good at Open Data Science East 2024
Generative AI for Social Good at Open Data Science East 2024Colleen Farrelly
 
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)jennyeacort
 
Identifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population MeanIdentifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population MeanMYRABACSAFRA2
 
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档208367051
 
DBA Basics: Getting Started with Performance Tuning.pdf
DBA Basics: Getting Started with Performance Tuning.pdfDBA Basics: Getting Started with Performance Tuning.pdf
DBA Basics: Getting Started with Performance Tuning.pdfJohn Sterrett
 
NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...
NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...
NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...Amil Baba Dawood bangali
 
Top 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In QueensTop 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In Queensdataanalyticsqueen03
 
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...Jack DiGiovanna
 
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改yuu sss
 
Call Girls In Dwarka 9654467111 Escorts Service
Call Girls In Dwarka 9654467111 Escorts ServiceCall Girls In Dwarka 9654467111 Escorts Service
Call Girls In Dwarka 9654467111 Escorts ServiceSapana Sha
 
RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.natarajan8993
 
Student Profile Sample report on improving academic performance by uniting gr...
Student Profile Sample report on improving academic performance by uniting gr...Student Profile Sample report on improving academic performance by uniting gr...
Student Profile Sample report on improving academic performance by uniting gr...Seán Kennedy
 
Machine learning classification ppt.ppt
Machine learning classification  ppt.pptMachine learning classification  ppt.ppt
Machine learning classification ppt.pptamreenkhanum0307
 
Heart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis ProjectHeart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis ProjectBoston Institute of Analytics
 
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort servicejennyeacort
 
20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdf20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdfHuman37
 
Predictive Analysis for Loan Default Presentation : Data Analysis Project PPT
Predictive Analysis for Loan Default  Presentation : Data Analysis Project PPTPredictive Analysis for Loan Default  Presentation : Data Analysis Project PPT
Predictive Analysis for Loan Default Presentation : Data Analysis Project PPTBoston Institute of Analytics
 
Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...
Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...
Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...ssuserf63bd7
 

Kürzlich hochgeladen (20)

Predicting Salary Using Data Science: A Comprehensive Analysis.pdf
Predicting Salary Using Data Science: A Comprehensive Analysis.pdfPredicting Salary Using Data Science: A Comprehensive Analysis.pdf
Predicting Salary Using Data Science: A Comprehensive Analysis.pdf
 
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
 
Generative AI for Social Good at Open Data Science East 2024
Generative AI for Social Good at Open Data Science East 2024Generative AI for Social Good at Open Data Science East 2024
Generative AI for Social Good at Open Data Science East 2024
 
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
Call Us ➥97111√47426🤳Call Girls in Aerocity (Delhi NCR)
 
Identifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population MeanIdentifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population Mean
 
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
 
DBA Basics: Getting Started with Performance Tuning.pdf
DBA Basics: Getting Started with Performance Tuning.pdfDBA Basics: Getting Started with Performance Tuning.pdf
DBA Basics: Getting Started with Performance Tuning.pdf
 
NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...
NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...
NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...
 
Top 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In QueensTop 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In Queens
 
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
 
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
 
Call Girls In Dwarka 9654467111 Escorts Service
Call Girls In Dwarka 9654467111 Escorts ServiceCall Girls In Dwarka 9654467111 Escorts Service
Call Girls In Dwarka 9654467111 Escorts Service
 
RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.
 
Student Profile Sample report on improving academic performance by uniting gr...
Student Profile Sample report on improving academic performance by uniting gr...Student Profile Sample report on improving academic performance by uniting gr...
Student Profile Sample report on improving academic performance by uniting gr...
 
Machine learning classification ppt.ppt
Machine learning classification  ppt.pptMachine learning classification  ppt.ppt
Machine learning classification ppt.ppt
 
Heart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis ProjectHeart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis Project
 
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
 
20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdf20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdf
 
Predictive Analysis for Loan Default Presentation : Data Analysis Project PPT
Predictive Analysis for Loan Default  Presentation : Data Analysis Project PPTPredictive Analysis for Loan Default  Presentation : Data Analysis Project PPT
Predictive Analysis for Loan Default Presentation : Data Analysis Project PPT
 
Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...
Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...
Statistics, Data Analysis, and Decision Modeling, 5th edition by James R. Eva...
 

A Human Perspective on Algorithmic Similarity

  • 1. A Human Perspective on Algorithmic Similarity RecSys 2020 Zach Schendel, Faraz Farzin, & Siddhi Sundar Netflix Product Innovation, Consumer Insights
  • 2. Even though it was highly recommended by a Netflix algorithm, I rejected this movie many times in 2017. It just didn’t look or sound good from the description.
  • 3. Importantly, I like to click on “More Like This” because I want to find out what unfamiliar movies are similar to. I have seen a few of these. They were decent, but the connection between them was not obvious
  • 4. I eventually watched it after reading more about it online. It remains my favorite Netflix Original movie! ...but, if I were to personalize “More Like This”, I would include similar titles that would be explanatory and authentic for me in that moment.
  • 5. And… I wasn’t the only one who had an opinion about similar recommendations on Netflix...
  • 6. These and other examples got us thinking… what exactly is similarity? How similar does something need to be to describe another movie? When do algorithms need to be restrictive, and when do they have permission to be relaxed? Where is the line between “similar” and “dissimilar”? We decided to ask Netflix members!
  • 7. We employed three different research methods. An international landscape assessment of how similarity is used in and outside of the Netflix category by other businesses. Qualitative interviews with Netflix members assessing how appropriate and successful our recommendations are in different places where similarity could be used as a driver for algorithms. A quantitative evaluation of perceived similarity between a series of source titles and potential recommendations using a technique called inverse Multi-dimensional Scaling (iMDS) [1]. More on this later... [1] Kriegeskorte & Mur, 2012 1. 2. 3.
  • 8. the Person Who is seeing the recommendations, and what is their past experience with the source? the Context What is going on at the moment? What are the user’s needs? 1 2 3the Placement Where are the recommendations placed within the Netflix user interface? The answer...it’s complicated! But we can really help our recommendations! There are 3 sources of complexity:
  • 9. the Person the Context 1 2 3the Placement Where are the recommendations placed within the Netflix user interface?
  • 10. Many places in the Netflix interface are populated by algorithms that factor in some dimension of similarity.
  • 11. We found that members had higher expectations of similarity when part of a 1:1 recommendation. but… this doesn’t make much sense at all. The Crown and The Office have little in common. This makes perfect sense. The Kissing Booth and To All the Boys... are teen romantic dramas with female leads. Both cases are risky - there are no backups/other titles to help explain the similarity.
  • 12. But, there were lower expectations in places with 1:many recommendations, like while browsing. Members are unlikely to say, “oh you liked Million Dollar Beach House? You should definitely watch Queer Eye.” ...but this broad placement is mostly made up of reality shows, so the link makes complete sense as a row.
  • 13. And, like earlier, there were higher expectations in placements that explicitly result from member action. If they search for something specific If they click into a specific title and navigate to “More Like This” Members are looking for something specifically similar.
  • 14. the Person the Context 1 2 3the Placement There is no one-size-fits all approach to employing similarity signals in algorithms in all candidate placements. Places that display 1:1 recommendations or react to explicit user engagement have higher expectations of similarity. Summary:
  • 15. the Placement the Context 1 2 3the Person Who is seeing the recommendations, and what is their past experience with the source?
  • 16. Method: We asked members to use iMDS[1] to self-cluster content by similarity. [1] Kriegeskorte & Mur, 2012 More similar = closer together. Will be added
  • 17. Let’s walk through one example to illustrate what we learned. Stranger Things is multidimensional. Sci-fi . Fantasy . Teen . Horror . Thriller . 80’s . Nostalgia . Starring Winona Ryder . Ominous . Scary . Exciting . Comedy . Coming of age
  • 18. These movies are very similar to Stranger Things on almost every one of those dimensions. These high-similarity options would work well in 1:1 placements discussed above. But, you quickly run out of options, and these won’t make up the bulk of placements.
  • 19. Both broad and specific drivers of perceived similarity can be used to fill out the bulk of placements. Sci-fi . Fantasy . Teen . Horror . Thriller . 80’s . Nostalgia . Starring Winona Ryder . Ominous . Scary . Exciting . Comedy . Coming of age
  • 20. Broad similarity drivers were surface level, like genre. They piqued interest and had clear source links. Sci-fi . Fantasy . Teen . Horror . Thriller . 80’s . Nostalgia . Starring Winona Ryder . Ominous . Scary . Exciting . Comedy . Coming of age
  • 21. Specific similarity drivers were varied and difficult to predict but were more salient final points of proof. it stars Winona Ryder! ‘80s nostalgia! Battling science monsters! Group of misfit teens in the 80’s!
  • 22. But...they are high-risk/high-reward. Trustbusters often occur when a wrong or unclear link are chosen. it stars Winona Ryder! ‘80s nostalgia! Battling science monsters! Group of misfit teens in the 80’s!
  • 23. the Person The degree to which something is or is not similar is in the eyes of the beholder, all of whom might latch on to different paths or give different permissions. The best path for algorithms is to find ways to balance broad and specific drivers of similarity. Broadly similar recommendations that differentially emphasize specific drivers for an individual should minimize trustbusters. the Context 1 2 3the Placement There is no one-size-fits all approach to employing similarity signals in algorithms in all candidate placements. Places that display 1:1 recommendations or react to explicit user engagement have higher expectations of similarity. Summary:
  • 24. the Placement the Person 1 2 3the Context What is going on at the moment? What are the user’s needs?
  • 25. Finally, to add even more complexity, even if you hold placement and person consistent, context also matters. After you finish a movie or show, Netflix recommends something.
  • 26. You finished... ...try this next. After finishing a show, members are most likely to watch something similar.
  • 27. 18%...watch another reality show #2 among options But while this is certainly a common path, it is not the most common or only path. 52%...watch another Netflix Original #1 among options One example… After finishing a Netflix Original reality TV show... 55%...watch more unserialized content #1 among options
  • 28. There are multiple contexts in which similarity is unnecessary and in fact, the opposite of ideal. “That was intense. I need a change of pace, something lighter” “I don’t have time for another movie. I’ll just put on something short.” … and more “I just need to rewatch something familiar, something I have watched before.”
  • 29. the Placement the Person 1 2 3the Context Even if the placement and the person are held constant, context further impacts the perception or need for similarity. Algos that can take context into consideration (e.g., today it’s likely to be more of the same day, tomorrow a change of pace) or allow for balance across multiple paths should be more successful. Summary:
  • 30. % Fewer Perceived Trustbusters comparing the old similarity model to the new similarity model by similarity rank. FEWERTrustbusters In the end, the research validated a new similarity model! There were fewer perceived trustbusters!
  • 31. And, for me at least, these are better recommendations! “More Like This” Similars
  • 32. Thanks! A Human Perspective on Algorithmic Similarity RecSys 2020 Zach Schendel, Faraz Farzin, & Siddhi Sundar Netflix Product Innovation, Consumer Insights
  • 33. Slides for 3 minute intro video A Human Perspective on Algorithmic Similarity RecSys 2020 Zach Schendel, Faraz Farzin, & Siddhi Sundar Netflix Product Innovation, Consumer Insights
  • 34. If you look at Twitter, there are typically 3 types of comments.
  • 35. I really liked this! … I want more like it!1
  • 36.
  • 38.
  • 40.
  • 41. “More Like This” Similars We are going to talk about how Netflix worked directly with member research to get from this?!?... ?!?
  • 42. ...to this!... “More Like This” Similars!
  • 43. the Person Who is seeing the recommendations, and what is their past experience with the source? the Context What is going on at the moment? What are the user’s needs? 1 2 3the Placement Where are the recommendations placed within the Netflix user interface? ...by exploring how perceived similarity can be strongly influenced by three new variables: