SlideShare una empresa de Scribd logo
1 de 27
©2016 LinkedIn Corporation. All Rights Reserved. 1
Audience Expansion for Online Social Network
Advertising
Haishan Liu, David Pardoe, Kun Liu, Manoj Thakur,
Frank Cao, Chongzhe Li
©2016 LinkedIn Corporation. All Rights Reserved. 2
Ad Targeting Overview
User / Ad
Matching
Ad campaign
geo = na.us || na.ca
&& title = software engineer
&& skill = java || C++
&& comp != LinkedIn
...
Member profile
geo: na.us.ny
title: data scientist
skill: java, hadoop
company: LinkedIn
...
Ad Campaign Member Profile
©2016 LinkedIn Corporation. All Rights Reserved. 3
What is Audience Expansion?
Increase the reach of ad campaigns
©2016 LinkedIn Corporation. All Rights Reserved. 4
Why Audience Expansion
▪ Enumerating targeting options
is hard
▪ Order of magnitude
of
common targeting attributes
▪ Title: 10k
▪ Skill: 100k
▪ Group: 1 million
▪ Company: 10 million
©2016 LinkedIn Corporation. All Rights Reserved. 5
Goal of Audience Expansion
▪ For advertisers
▪ Increased reach
▪ Simplified targeting
▪ For LinkedIn
▪ Increased utilization of inventory
▪ Higher marketplace participation
▪ For members
▪ Minimal impact on experience
©2016 LinkedIn Corporation. All Rights Reserved. 6
More targeting options (groups, skills, schools, etc.)
©2016 LinkedIn Corporation. All Rights Reserved. 7
Audience Expansion Methodology Overview
● Two types of expansion work simultaneously
○ Campaign-agnostic expansion
■ Matches targeting criteria with expanded member profile
attributes
■ Takes effect immediately after the campaign launch
○ Campaign-aware expansion
■ Matches a campaign to members similar to the targeted audience
■ The expanded audience is calculated offline, which takes a few
hours
©2016 LinkedIn Corporation. All Rights Reserved. 8
Audience
EXACT MATCH BROADENED MATCH
- Company: Nationwide Children's Hospital
- Title: Research Associate
- Skill: Comparative genomics, BioStats
Standard Profile Expanded Profile
- Company: St. Jude Children's Hospital
- Title: Research Scientist
- Skill: Bioinformatics
Campaign-Agnostic Expansion
©2016 LinkedIn Corporation. All Rights Reserved. 9
Audience
EXACT MATCH LOOK-ALIKE AUDIENCE
- Profile similarity
- Behavior similarity
Campaign-Aware Expansion
©2016 LinkedIn Corporation. All Rights Reserved. 10
Audience Expansion in the Ads Serving Flow
©2016 LinkedIn Corporation. All Rights Reserved. 11
Profile Attribute Expansion by Similar-X
Profile company: LinkedIn Profile group
… Other profile attribute
expansionsFor each member
Campaign-agnostic expansion
Similar companies Similar groups
Facebook
Twitter
Google
Pinterest
Netflix
Amazon
Amazon
Salesforce
Expanded profile:
©2016 LinkedIn Corporation. All Rights Reserved. 12
Modeling Overview for Similar-X
Entity similarity
Field similarity
©2016 LinkedIn Corporation. All Rights Reserved. 13
Feature Extraction (E.g., LinkedIn)
▪ N-grams/phrases
▪ headline: Internet, professional, Social Network
▪ description: connection, productive, Talent Solution
▪ Standardized
▪ industry: Internet
▪ type: public company
▪ size: 5001-10000 employees
©2016 LinkedIn Corporation. All Rights Reserved. 14
Feature Extraction (E.g., LinkedIn)
▪ Derived
▪ skills: Software Engineering, Management, Marketing
▪ interests: professional identity, software development
▪ Network proximities
▪ view-browsemap: Facebook, Twitter, Pinterest
▪ occupation-browsemap: Google, Yahoo, Facebook
©2016 LinkedIn Corporation. All Rights Reserved. 15
Lookalike Expansion by Similar-Profiles
originally targeted members
For each campaign
Campaign-aware expansion
SWE at LinkedIn
with a Machine
Learning skill
Data Scientist at
Amazon with a
Data Mining skill
©2016 LinkedIn Corporation. All Rights Reserved. 16
Similar Profiles
▪ Problem size the square of
LinkedIn users
▪ Serves as an independent
product in LinkedIn Recruiter
▪ Enables Search By Ideal
Candidate as part of Next-Gen
Recruiter
▪ Powers Lead
Recommendations in Sales
Navigator
©2016 LinkedIn Corporation. All Rights Reserved. 17
Other Considerations - Campaign Selection
▪ A set of heuristics to identify campaigns that would likely to benefit
from campaign-aware expansion
▪ Must target a relatively specific audience using fine-grained attributes
▪ The size of the audience should not be extremely large
▪ The budget should be enough to afford additional impressions from
the expanded audience
©2016 LinkedIn Corporation. All Rights Reserved. 18
Other Considerations - Offline Matching
▪ Offline pipeline to emulate online ad/member matching
▪ Matches all active campaigns against all members (~460 million)
▪ Implemented as a massive map-reduce job with following
optimizations
▪ Use inverted campaign index as opposed to forward campaign index
used in online matching
▪ Use distributed cache to replicate campaign inverted index across all
mappers for matching
▪ Reduce the number of campaigns selected for matching by
considering only active campaigns
©2016 LinkedIn Corporation. All Rights Reserved. 19
Other Considerations - Post-Expansion Filters
▪ Remove following members from the expanded audience
▪ Negatively targeted
▪ Those violate the location, gender, or language criteria
▪ In addition, remove most expensive members
▪ Estimate member eCPC
▪ Regress historical bid distribution on profile attributes
©2016 LinkedIn Corporation. All Rights Reserved. 20
Other Considerations - Online-Offline Sync
Tc = -1
Tg = t1
campaign-aware
pipeline run
campaign
targeting
change
Tc = t2
Tg = t1
Tc > Tg
t1 t2 t3
Tc = t3
Tg = t1
Tc > Tg
Campaign targeting could change after the offline computation
time
campaign-aware
pipeline run
Offline results generated before t2 will be
invalidated until new results are generated after t3
©2016 LinkedIn Corporation. All Rights Reserved. 21
Effectiveness of Similar Profiles
▪ Compare network density
between original and expanded
audiences
▪ Baseline -- expand to active ads
clickers
▪ Expansion by Similar Profiles
achieves much higher network
density than baseline
denseraudience
more expansion
©2016 LinkedIn Corporation. All Rights Reserved. 22
Campaign-Member Fitness Score
▪ Campaign-member score
▪ The more similar members there
are in the targeted audience to an
untargeted member, the more fit
the untargeted member is to be
included in the expanded
audience.
©2016 LinkedIn Corporation. All Rights Reserved. 23
Evaluation and Results
Tuning a number of hyper-parameters through multiple A/B tests on a
fraction of members
©2016 LinkedIn Corporation. All Rights Reserved. 24
Online Evaluation
Relative changes in metrics compared to control. Statistically significant differences are in bold.
Relative changes in metrics when only considering impressions won by expansion-enabled campaigns.
©2016 LinkedIn Corporation. All Rights Reserved. 25
Impact of Audience Expansion to the Marketplace
Coverage: the number of delivered impressions as a percentage of the number of available
impressions, which measures how effective the ads inventory is utilized
Coverage increases in response to the Audience Expansion ramp-up
©2016 LinkedIn Corporation. All Rights Reserved. 26
Future Work
▪ Beyond Similar-X
▪ behavior-based similarity (ads click/share/like)
▪ collaborative-filtering
▪ Beyond targeting-based expansion
▪ Expand from optional seed lists
©2016 LinkedIn Corporation. All Rights Reserved. 27
Thanks For Your Attention!
Questions?

Más contenido relacionado

Destacado

Internet Advertising With Exact Drive
Internet Advertising With Exact DriveInternet Advertising With Exact Drive
Internet Advertising With Exact DriveExactDrive Inc.
 
APG NL Presentation by Heather LeFevre
APG NL Presentation by Heather LeFevreAPG NL Presentation by Heather LeFevre
APG NL Presentation by Heather LeFevreHeather LeFevre
 
Chipotle persona profile by Iliana Torres-Sanchez
Chipotle persona profile by Iliana Torres-Sanchez Chipotle persona profile by Iliana Torres-Sanchez
Chipotle persona profile by Iliana Torres-Sanchez Iliana t
 
What Gardening Can Teach Us
What Gardening Can Teach UsWhat Gardening Can Teach Us
What Gardening Can Teach UsHeather LeFevre
 
Apresentação Trabalho Conclusão de Curso Rebranding Lojas Renner
Apresentação Trabalho Conclusão de Curso Rebranding Lojas RennerApresentação Trabalho Conclusão de Curso Rebranding Lojas Renner
Apresentação Trabalho Conclusão de Curso Rebranding Lojas RennerJonathan Fabris
 
Target Audience Analysis & Archetypes (Outdoor Technical Clothing)
Target Audience Analysis & Archetypes (Outdoor Technical Clothing)Target Audience Analysis & Archetypes (Outdoor Technical Clothing)
Target Audience Analysis & Archetypes (Outdoor Technical Clothing)Marco Gervasio
 
How to classify the visual space of a brand identity
How to classify the visual space of a brand identityHow to classify the visual space of a brand identity
How to classify the visual space of a brand identityEducreate
 
Effective search and social in 2016 | Online advertising | South West Network...
Effective search and social in 2016 | Online advertising | South West Network...Effective search and social in 2016 | Online advertising | South West Network...
Effective search and social in 2016 | Online advertising | South West Network...CharityComms
 
Decoding Resurgent Pride in American Origin
Decoding Resurgent Pride in American OriginDecoding Resurgent Pride in American Origin
Decoding Resurgent Pride in American OriginJohn W. Manley
 

Destacado (13)

Internet Advertising With Exact Drive
Internet Advertising With Exact DriveInternet Advertising With Exact Drive
Internet Advertising With Exact Drive
 
Propagation Planning
Propagation PlanningPropagation Planning
Propagation Planning
 
APG NL Presentation by Heather LeFevre
APG NL Presentation by Heather LeFevreAPG NL Presentation by Heather LeFevre
APG NL Presentation by Heather LeFevre
 
Chipotle persona profile by Iliana Torres-Sanchez
Chipotle persona profile by Iliana Torres-Sanchez Chipotle persona profile by Iliana Torres-Sanchez
Chipotle persona profile by Iliana Torres-Sanchez
 
The Planner Survey 2009
The Planner  Survey  2009The Planner  Survey  2009
The Planner Survey 2009
 
What Gardening Can Teach Us
What Gardening Can Teach UsWhat Gardening Can Teach Us
What Gardening Can Teach Us
 
Apresentação Trabalho Conclusão de Curso Rebranding Lojas Renner
Apresentação Trabalho Conclusão de Curso Rebranding Lojas RennerApresentação Trabalho Conclusão de Curso Rebranding Lojas Renner
Apresentação Trabalho Conclusão de Curso Rebranding Lojas Renner
 
Target Audience Analysis & Archetypes (Outdoor Technical Clothing)
Target Audience Analysis & Archetypes (Outdoor Technical Clothing)Target Audience Analysis & Archetypes (Outdoor Technical Clothing)
Target Audience Analysis & Archetypes (Outdoor Technical Clothing)
 
How to classify the visual space of a brand identity
How to classify the visual space of a brand identityHow to classify the visual space of a brand identity
How to classify the visual space of a brand identity
 
Effective search and social in 2016 | Online advertising | South West Network...
Effective search and social in 2016 | Online advertising | South West Network...Effective search and social in 2016 | Online advertising | South West Network...
Effective search and social in 2016 | Online advertising | South West Network...
 
Stopping the Dropout
Stopping the DropoutStopping the Dropout
Stopping the Dropout
 
Declare Your Minor
Declare Your MinorDeclare Your Minor
Declare Your Minor
 
Decoding Resurgent Pride in American Origin
Decoding Resurgent Pride in American OriginDecoding Resurgent Pride in American Origin
Decoding Resurgent Pride in American Origin
 

Similar a Audience Expansion for Online Social Network Advertising - KDD 2016

Lead Generation
Lead GenerationLead Generation
Lead GenerationAmy Smith
 
Webinar Display advertising secrets for scale
Webinar Display advertising secrets for scaleWebinar Display advertising secrets for scale
Webinar Display advertising secrets for scaleMichael O'Connell
 
Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...
Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...
Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...e-dialog GmbH
 
We dig tech startups - Gravytrain '15
We dig tech startups   - Gravytrain '15We dig tech startups   - Gravytrain '15
We dig tech startups - Gravytrain '15Dan Whitehouse
 
Masterclass: Exceed Your Demand Generation Goals
Masterclass: Exceed Your Demand Generation GoalsMasterclass: Exceed Your Demand Generation Goals
Masterclass: Exceed Your Demand Generation GoalsLinkedIn
 
Maximize Performance of Your Campaigns with Sponsored Updates Partners
Maximize Performance of Your Campaigns with Sponsored Updates PartnersMaximize Performance of Your Campaigns with Sponsored Updates Partners
Maximize Performance of Your Campaigns with Sponsored Updates PartnersLinkedIn
 
Audience Growth Bootcamp - Growth solutions
Audience Growth Bootcamp - Growth solutionsAudience Growth Bootcamp - Growth solutions
Audience Growth Bootcamp - Growth solutionsMSFTAdvertising
 
LinkedIn Masterclass January 2016
LinkedIn Masterclass January 2016LinkedIn Masterclass January 2016
LinkedIn Masterclass January 2016GlowMetrics
 
Live Webinar: Getting Started wit Display On and Off LinkedIn
Live Webinar: Getting Started wit Display On and Off LinkedInLive Webinar: Getting Started wit Display On and Off LinkedIn
Live Webinar: Getting Started wit Display On and Off LinkedInLinkedIn
 
How to Optimise the Efficiency of Your Digital Communication
How to Optimise the Efficiency of Your Digital CommunicationHow to Optimise the Efficiency of Your Digital Communication
How to Optimise the Efficiency of Your Digital CommunicationLa French Tech
 
LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015
LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015
LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015LinkedIn
 
PYLON for LinkedIn Engagement Insights
PYLON for LinkedIn Engagement InsightsPYLON for LinkedIn Engagement Insights
PYLON for LinkedIn Engagement InsightsLinkedIn
 
How to run an effective linked in campaign
How to run an effective linked in campaignHow to run an effective linked in campaign
How to run an effective linked in campaignteam-abr
 
In the Lab with DeeBee: Orchestration 101.pdf
In the Lab with DeeBee: Orchestration 101.pdfIn the Lab with DeeBee: Orchestration 101.pdf
In the Lab with DeeBee: Orchestration 101.pdfDemandbase
 
Mastering B2B Advertising on LinkedIn: Expert Campaign Optimization Tips
Mastering B2B Advertising on LinkedIn: Expert Campaign Optimization TipsMastering B2B Advertising on LinkedIn: Expert Campaign Optimization Tips
Mastering B2B Advertising on LinkedIn: Expert Campaign Optimization TipsSevenAtoms Inc.
 
Webinar: Retargeting to the Max
Webinar: Retargeting to the MaxWebinar: Retargeting to the Max
Webinar: Retargeting to the MaxKatana Media
 
LinkedIn for tech marketers
LinkedIn for tech marketersLinkedIn for tech marketers
LinkedIn for tech marketersTony Weinberg
 
Innovating digital, 2019 and beyond - CIM West Yorkshire from Dave Chaffey
Innovating digital, 2019 and beyond - CIM West Yorkshire from Dave ChaffeyInnovating digital, 2019 and beyond - CIM West Yorkshire from Dave Chaffey
Innovating digital, 2019 and beyond - CIM West Yorkshire from Dave ChaffeyDave Chaffey
 

Similar a Audience Expansion for Online Social Network Advertising - KDD 2016 (20)

Lead Generation
Lead GenerationLead Generation
Lead Generation
 
Webinar Display advertising secrets for scale
Webinar Display advertising secrets for scaleWebinar Display advertising secrets for scale
Webinar Display advertising secrets for scale
 
Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...
Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...
Google Analytics Konferenz 2018_Linking DoubleClick & GA 360_Felix Derkum (Go...
 
We dig tech startups - Gravytrain '15
We dig tech startups   - Gravytrain '15We dig tech startups   - Gravytrain '15
We dig tech startups - Gravytrain '15
 
Kenshoo Social Presentation
Kenshoo Social PresentationKenshoo Social Presentation
Kenshoo Social Presentation
 
Masterclass: Exceed Your Demand Generation Goals
Masterclass: Exceed Your Demand Generation GoalsMasterclass: Exceed Your Demand Generation Goals
Masterclass: Exceed Your Demand Generation Goals
 
Maximize Performance of Your Campaigns with Sponsored Updates Partners
Maximize Performance of Your Campaigns with Sponsored Updates PartnersMaximize Performance of Your Campaigns with Sponsored Updates Partners
Maximize Performance of Your Campaigns with Sponsored Updates Partners
 
Audience Growth Bootcamp - Growth solutions
Audience Growth Bootcamp - Growth solutionsAudience Growth Bootcamp - Growth solutions
Audience Growth Bootcamp - Growth solutions
 
LinkedIn Masterclass January 2016
LinkedIn Masterclass January 2016LinkedIn Masterclass January 2016
LinkedIn Masterclass January 2016
 
Live Webinar: Getting Started wit Display On and Off LinkedIn
Live Webinar: Getting Started wit Display On and Off LinkedInLive Webinar: Getting Started wit Display On and Off LinkedIn
Live Webinar: Getting Started wit Display On and Off LinkedIn
 
How to Optimise the Efficiency of Your Digital Communication
How to Optimise the Efficiency of Your Digital CommunicationHow to Optimise the Efficiency of Your Digital Communication
How to Optimise the Efficiency of Your Digital Communication
 
LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015
LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015
LinkedIn Lead Accelerator & Network Display - B2B Connect SP 2015
 
NYC HUG 2019: Advertising on LinkedIn
NYC HUG 2019: Advertising on LinkedInNYC HUG 2019: Advertising on LinkedIn
NYC HUG 2019: Advertising on LinkedIn
 
PYLON for LinkedIn Engagement Insights
PYLON for LinkedIn Engagement InsightsPYLON for LinkedIn Engagement Insights
PYLON for LinkedIn Engagement Insights
 
How to run an effective linked in campaign
How to run an effective linked in campaignHow to run an effective linked in campaign
How to run an effective linked in campaign
 
In the Lab with DeeBee: Orchestration 101.pdf
In the Lab with DeeBee: Orchestration 101.pdfIn the Lab with DeeBee: Orchestration 101.pdf
In the Lab with DeeBee: Orchestration 101.pdf
 
Mastering B2B Advertising on LinkedIn: Expert Campaign Optimization Tips
Mastering B2B Advertising on LinkedIn: Expert Campaign Optimization TipsMastering B2B Advertising on LinkedIn: Expert Campaign Optimization Tips
Mastering B2B Advertising on LinkedIn: Expert Campaign Optimization Tips
 
Webinar: Retargeting to the Max
Webinar: Retargeting to the MaxWebinar: Retargeting to the Max
Webinar: Retargeting to the Max
 
LinkedIn for tech marketers
LinkedIn for tech marketersLinkedIn for tech marketers
LinkedIn for tech marketers
 
Innovating digital, 2019 and beyond - CIM West Yorkshire from Dave Chaffey
Innovating digital, 2019 and beyond - CIM West Yorkshire from Dave ChaffeyInnovating digital, 2019 and beyond - CIM West Yorkshire from Dave Chaffey
Innovating digital, 2019 and beyond - CIM West Yorkshire from Dave Chaffey
 

Último

20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdf20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdfHuman37
 
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
 
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
 
办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一
办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一
办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一F sss
 
Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...
Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...
Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...Boston Institute of Analytics
 
Easter Eggs From Star Wars and in cars 1 and 2
Easter Eggs From Star Wars and in cars 1 and 2Easter Eggs From Star Wars and in cars 1 and 2
Easter Eggs From Star Wars and in cars 1 and 217djon017
 
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
 
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
 
RadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.pdfRadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.pdfgstagge
 
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
 
Multiple time frame trading analysis -brianshannon.pdf
Multiple time frame trading analysis -brianshannon.pdfMultiple time frame trading analysis -brianshannon.pdf
Multiple time frame trading analysis -brianshannon.pdfchwongval
 
LLMs, LMMs, their Improvement Suggestions and the Path towards AGI
LLMs, LMMs, their Improvement Suggestions and the Path towards AGILLMs, LMMs, their Improvement Suggestions and the Path towards AGI
LLMs, LMMs, their Improvement Suggestions and the Path towards AGIThomas Poetter
 
INTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTDINTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTDRafezzaman
 
Semantic Shed - Squashing and Squeezing.pptx
Semantic Shed - Squashing and Squeezing.pptxSemantic Shed - Squashing and Squeezing.pptx
Semantic Shed - Squashing and Squeezing.pptxMike Bennett
 
Conf42-LLM_Adding Generative AI to Real-Time Streaming Pipelines
Conf42-LLM_Adding Generative AI to Real-Time Streaming PipelinesConf42-LLM_Adding Generative AI to Real-Time Streaming Pipelines
Conf42-LLM_Adding Generative AI to Real-Time Streaming PipelinesTimothy Spann
 
科罗拉多大学波尔得分校毕业证学位证成绩单-可办理
科罗拉多大学波尔得分校毕业证学位证成绩单-可办理科罗拉多大学波尔得分校毕业证学位证成绩单-可办理
科罗拉多大学波尔得分校毕业证学位证成绩单-可办理e4aez8ss
 
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝DelhiRS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhijennyeacort
 
Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...
Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...
Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...Thomas Poetter
 
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
 
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
 

Último (20)

20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdf20240419 - Measurecamp Amsterdam - SAM.pdf
20240419 - Measurecamp Amsterdam - SAM.pdf
 
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
 
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...
 
办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一
办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一
办理学位证中佛罗里达大学毕业证,UCF成绩单原版一比一
 
Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...
Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...
Data Analysis Project : Targeting the Right Customers, Presentation on Bank M...
 
Easter Eggs From Star Wars and in cars 1 and 2
Easter Eggs From Star Wars and in cars 1 and 2Easter Eggs From Star Wars and in cars 1 and 2
Easter Eggs From Star Wars and in cars 1 and 2
 
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
 
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
 
RadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.pdfRadioAdProWritingCinderellabyButleri.pdf
RadioAdProWritingCinderellabyButleri.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
 
Multiple time frame trading analysis -brianshannon.pdf
Multiple time frame trading analysis -brianshannon.pdfMultiple time frame trading analysis -brianshannon.pdf
Multiple time frame trading analysis -brianshannon.pdf
 
LLMs, LMMs, their Improvement Suggestions and the Path towards AGI
LLMs, LMMs, their Improvement Suggestions and the Path towards AGILLMs, LMMs, their Improvement Suggestions and the Path towards AGI
LLMs, LMMs, their Improvement Suggestions and the Path towards AGI
 
INTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTDINTERNSHIP ON PURBASHA COMPOSITE TEX LTD
INTERNSHIP ON PURBASHA COMPOSITE TEX LTD
 
Semantic Shed - Squashing and Squeezing.pptx
Semantic Shed - Squashing and Squeezing.pptxSemantic Shed - Squashing and Squeezing.pptx
Semantic Shed - Squashing and Squeezing.pptx
 
Conf42-LLM_Adding Generative AI to Real-Time Streaming Pipelines
Conf42-LLM_Adding Generative AI to Real-Time Streaming PipelinesConf42-LLM_Adding Generative AI to Real-Time Streaming Pipelines
Conf42-LLM_Adding Generative AI to Real-Time Streaming Pipelines
 
科罗拉多大学波尔得分校毕业证学位证成绩单-可办理
科罗拉多大学波尔得分校毕业证学位证成绩单-可办理科罗拉多大学波尔得分校毕业证学位证成绩单-可办理
科罗拉多大学波尔得分校毕业证学位证成绩单-可办理
 
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝DelhiRS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
 
Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...
Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...
Minimizing AI Hallucinations/Confabulations and the Path towards AGI with Exa...
 
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
 
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
 

Audience Expansion for Online Social Network Advertising - KDD 2016

  • 1. ©2016 LinkedIn Corporation. All Rights Reserved. 1 Audience Expansion for Online Social Network Advertising Haishan Liu, David Pardoe, Kun Liu, Manoj Thakur, Frank Cao, Chongzhe Li
  • 2. ©2016 LinkedIn Corporation. All Rights Reserved. 2 Ad Targeting Overview User / Ad Matching Ad campaign geo = na.us || na.ca && title = software engineer && skill = java || C++ && comp != LinkedIn ... Member profile geo: na.us.ny title: data scientist skill: java, hadoop company: LinkedIn ... Ad Campaign Member Profile
  • 3. ©2016 LinkedIn Corporation. All Rights Reserved. 3 What is Audience Expansion? Increase the reach of ad campaigns
  • 4. ©2016 LinkedIn Corporation. All Rights Reserved. 4 Why Audience Expansion ▪ Enumerating targeting options is hard ▪ Order of magnitude of common targeting attributes ▪ Title: 10k ▪ Skill: 100k ▪ Group: 1 million ▪ Company: 10 million
  • 5. ©2016 LinkedIn Corporation. All Rights Reserved. 5 Goal of Audience Expansion ▪ For advertisers ▪ Increased reach ▪ Simplified targeting ▪ For LinkedIn ▪ Increased utilization of inventory ▪ Higher marketplace participation ▪ For members ▪ Minimal impact on experience
  • 6. ©2016 LinkedIn Corporation. All Rights Reserved. 6 More targeting options (groups, skills, schools, etc.)
  • 7. ©2016 LinkedIn Corporation. All Rights Reserved. 7 Audience Expansion Methodology Overview ● Two types of expansion work simultaneously ○ Campaign-agnostic expansion ■ Matches targeting criteria with expanded member profile attributes ■ Takes effect immediately after the campaign launch ○ Campaign-aware expansion ■ Matches a campaign to members similar to the targeted audience ■ The expanded audience is calculated offline, which takes a few hours
  • 8. ©2016 LinkedIn Corporation. All Rights Reserved. 8 Audience EXACT MATCH BROADENED MATCH - Company: Nationwide Children's Hospital - Title: Research Associate - Skill: Comparative genomics, BioStats Standard Profile Expanded Profile - Company: St. Jude Children's Hospital - Title: Research Scientist - Skill: Bioinformatics Campaign-Agnostic Expansion
  • 9. ©2016 LinkedIn Corporation. All Rights Reserved. 9 Audience EXACT MATCH LOOK-ALIKE AUDIENCE - Profile similarity - Behavior similarity Campaign-Aware Expansion
  • 10. ©2016 LinkedIn Corporation. All Rights Reserved. 10 Audience Expansion in the Ads Serving Flow
  • 11. ©2016 LinkedIn Corporation. All Rights Reserved. 11 Profile Attribute Expansion by Similar-X Profile company: LinkedIn Profile group … Other profile attribute expansionsFor each member Campaign-agnostic expansion Similar companies Similar groups Facebook Twitter Google Pinterest Netflix Amazon Amazon Salesforce Expanded profile:
  • 12. ©2016 LinkedIn Corporation. All Rights Reserved. 12 Modeling Overview for Similar-X Entity similarity Field similarity
  • 13. ©2016 LinkedIn Corporation. All Rights Reserved. 13 Feature Extraction (E.g., LinkedIn) ▪ N-grams/phrases ▪ headline: Internet, professional, Social Network ▪ description: connection, productive, Talent Solution ▪ Standardized ▪ industry: Internet ▪ type: public company ▪ size: 5001-10000 employees
  • 14. ©2016 LinkedIn Corporation. All Rights Reserved. 14 Feature Extraction (E.g., LinkedIn) ▪ Derived ▪ skills: Software Engineering, Management, Marketing ▪ interests: professional identity, software development ▪ Network proximities ▪ view-browsemap: Facebook, Twitter, Pinterest ▪ occupation-browsemap: Google, Yahoo, Facebook
  • 15. ©2016 LinkedIn Corporation. All Rights Reserved. 15 Lookalike Expansion by Similar-Profiles originally targeted members For each campaign Campaign-aware expansion SWE at LinkedIn with a Machine Learning skill Data Scientist at Amazon with a Data Mining skill
  • 16. ©2016 LinkedIn Corporation. All Rights Reserved. 16 Similar Profiles ▪ Problem size the square of LinkedIn users ▪ Serves as an independent product in LinkedIn Recruiter ▪ Enables Search By Ideal Candidate as part of Next-Gen Recruiter ▪ Powers Lead Recommendations in Sales Navigator
  • 17. ©2016 LinkedIn Corporation. All Rights Reserved. 17 Other Considerations - Campaign Selection ▪ A set of heuristics to identify campaigns that would likely to benefit from campaign-aware expansion ▪ Must target a relatively specific audience using fine-grained attributes ▪ The size of the audience should not be extremely large ▪ The budget should be enough to afford additional impressions from the expanded audience
  • 18. ©2016 LinkedIn Corporation. All Rights Reserved. 18 Other Considerations - Offline Matching ▪ Offline pipeline to emulate online ad/member matching ▪ Matches all active campaigns against all members (~460 million) ▪ Implemented as a massive map-reduce job with following optimizations ▪ Use inverted campaign index as opposed to forward campaign index used in online matching ▪ Use distributed cache to replicate campaign inverted index across all mappers for matching ▪ Reduce the number of campaigns selected for matching by considering only active campaigns
  • 19. ©2016 LinkedIn Corporation. All Rights Reserved. 19 Other Considerations - Post-Expansion Filters ▪ Remove following members from the expanded audience ▪ Negatively targeted ▪ Those violate the location, gender, or language criteria ▪ In addition, remove most expensive members ▪ Estimate member eCPC ▪ Regress historical bid distribution on profile attributes
  • 20. ©2016 LinkedIn Corporation. All Rights Reserved. 20 Other Considerations - Online-Offline Sync Tc = -1 Tg = t1 campaign-aware pipeline run campaign targeting change Tc = t2 Tg = t1 Tc > Tg t1 t2 t3 Tc = t3 Tg = t1 Tc > Tg Campaign targeting could change after the offline computation time campaign-aware pipeline run Offline results generated before t2 will be invalidated until new results are generated after t3
  • 21. ©2016 LinkedIn Corporation. All Rights Reserved. 21 Effectiveness of Similar Profiles ▪ Compare network density between original and expanded audiences ▪ Baseline -- expand to active ads clickers ▪ Expansion by Similar Profiles achieves much higher network density than baseline denseraudience more expansion
  • 22. ©2016 LinkedIn Corporation. All Rights Reserved. 22 Campaign-Member Fitness Score ▪ Campaign-member score ▪ The more similar members there are in the targeted audience to an untargeted member, the more fit the untargeted member is to be included in the expanded audience.
  • 23. ©2016 LinkedIn Corporation. All Rights Reserved. 23 Evaluation and Results Tuning a number of hyper-parameters through multiple A/B tests on a fraction of members
  • 24. ©2016 LinkedIn Corporation. All Rights Reserved. 24 Online Evaluation Relative changes in metrics compared to control. Statistically significant differences are in bold. Relative changes in metrics when only considering impressions won by expansion-enabled campaigns.
  • 25. ©2016 LinkedIn Corporation. All Rights Reserved. 25 Impact of Audience Expansion to the Marketplace Coverage: the number of delivered impressions as a percentage of the number of available impressions, which measures how effective the ads inventory is utilized Coverage increases in response to the Audience Expansion ramp-up
  • 26. ©2016 LinkedIn Corporation. All Rights Reserved. 26 Future Work ▪ Beyond Similar-X ▪ behavior-based similarity (ads click/share/like) ▪ collaborative-filtering ▪ Beyond targeting-based expansion ▪ Expand from optional seed lists
  • 27. ©2016 LinkedIn Corporation. All Rights Reserved. 27 Thanks For Your Attention! Questions?