SlideShare a Scribd company logo
1 of 21
A 10-Month-Long Deployment Study of
On-Demand Recruiting for
Low-Latency Crowdsourcing
Live Feedback & Questions: http://tinyurl.com/20171026IGNITION
Ting-Hao (Kenneth) Huang
Jeffrey P. Bigham
Carnegie Mellon University
Image: https://pixabay.com/photo-743920/
2
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Chorus: A Crowd-powered
Conversation Assistant
"Is there anything else I can help you with?": Challenges in Deploying an On-Demand Crowd-Powered
Conversational Agent. Huang et. al., HCOMP 2016
3
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Constraints of Deploying Chorus
• Small user traffic
• Avg. 2.97 conversations per day
• Fast worker response time
• Response time to a message ~= 32 sec (student participant)*
• Response time to a message ~= 105 sec (startup participant)*
• Need multiple workers to work together
• Limited budget
• 24-7 service
* IM waiting: timing and responsiveness in semi-synchronous communication. Avrahami et al. CSCW’08.
…Similar to many academic projects !
4
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
How to recruit workers
quickly on-demand?
On-demand
(QuikTurkit,
Bigham, et al., ‘10)
On-call
(Retainer Model,
Bernstein, et al.,‘11)
Ignition
Method Post HITs + SEO Retainer Hybrid
Latency 60 - 120 seconds < 10 seconds 20 – 90 seconds
How to Hold
Workers
Using old tasks Waiting Page Waiting Page
Recruiting Cost
Per Task
Old tasks
+ SEO (mturk fee)
All Workers’
Waiting Time
55% of Workers’
Waiting Time
Recruiting Cost
For Small
Deployments
Very Low
( k* #tasks )
High
( k* duration )
Low
( k* #tasks +)
…Can we have
something
inexpensive
and fast?
5
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Engines Generate Power
On-demand
(QuikTurkit,
Bigham, et al., ‘10)
On-call
(Retainer Model,
Bernstein, et al.,‘11)
Ignition
Method Post HITs + SEO Retainer Hybrid
Latency 60 - 120 seconds < 10 seconds 20 – 90 seconds
How to Hold
Workers
Using old tasks Waiting Page Waiting Page
Recruiting Cost
Per Task
Old tasks
+ SEO (mturk fee)
All Workers’
Waiting Time
55% of Workers’
Waiting Time
Recruiting Cost
For Small
Deployments
Very Low
( k* #tasks )
High
( k* duration )
Low
( k* #tasks +)
• Crowd-powered System
Why not
both?
Re-Start
the
Engine
Each Time
Start the
Engine
Once, and
Hold the
Power
Image: https://commons.wikimedia.org/wiki/File:3791_cc_Holden_3800_V6_engine_(2015-08-29)_03.jpg
6
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Ignition Model
1. Hire slightly more workers than we need
2. Hold the extra workers in the retainer
3. Always call workers from retainer first, and then post HITs
7
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Ignition = Retainer + Market
• When you want to recruit 8 workers each time…
• A Chorus task need at most 5 workers
#Workers
Recruited from
Retainer
#Workers Recruited from
Mturk Market
( Posting a HIT w/
Cmarket Assignments )
Posting More Assignments -> Gets Workers Faster
Cretainer + Cmarket = 8
8
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Waiting Page & Point System
http://TinyUrl.com/IgnitionDemo
9
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Transition Graph for Workers in Ignition
10
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Transition Graph for Workers in Ignition
11
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Transition Graph for Workers in Ignition
12
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
1. The task occurs instantly [ Instant ]
2. The task occurs later [ Retained ]
3. No tasks at all [ No Task ]
Transition Graph for Workers in Ignition
13
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Ignition Deployment
• From July 1st, 2016 to April 27th, 2017 (~10 months)
• 6,823 assignments, 648 workers
• $0.25 for a 30-minute waiting time
• Mturk Qualifications Used
• HIT Approval Rate > 90%, #Approved HITs> 200, Adult Content Qualification
Assignments Tasks (Conversation)
#workers in retainer
50% of tasks occurred when
1+ workers in the retainer
14
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Posting More Assignments -> Gets Workers Faster
Simply Posting HITs
Getting 1+ Workers Getting 3+ Workers
X-Axis: Time (x seconds)
Y-Axis: Prob(Getting N+ workers within x seconds)
Cmarket = 3
Cmarket
= 8
Cmarket = 8
Cmarket = 3
15
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Simply Posting HITs v.s. Ignition
Getting 1+ Workers Getting 3+ Workers
Cmarket
= 8
Cmarket = 8
Cmarket = 8
Cmarket = 8
16
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Does a Deployed Retainer Work?
• Yes, and (surprisingly) stable.
• Response rate = Proportion of workers in the retainer
who responded within 20 secs when being called
Response rate is stable over time.
Month of a Year Day of a Week Retainer Time
17
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
What do workers think?
• Worker Survey
• Collect opinions and self-reported behavior
• Ask 156 workers who completed 10+ Ignition HITs
• $2 compensation (~10mins)
• 101 workers finished the survey (Response Rate = 64.74%)
18
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
General HITs v.s. Ignition HITs
"I like doing HITs on MTurk in general." "I like doing Retainer HITs."
• 5-point Likert scale
• Strongly Disagree to Strongly Agree
19
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
How to recruit workers
quickly on-demand?
On-demand
(QuikTurkit,
Bigham, et al., ‘10)
On-call
(Retainer Model,
Bernstein, et al.,‘11)
Ignition
Method Post HITs + SEO Retainer Hybrid
Latency 60 - 120 seconds < 10 seconds 20 – 90 seconds
How to Hold
Workers
Using old tasks Waiting Page Waiting Page
Recruiting Cost
Per Task
Old tasks
+ SEO (mturk fee)
All Workers’
Waiting Time
55% of Workers’
Waiting Time
Recruiting Cost
For Small
Deployment
Very Low
( k* #tasks )
High
( k* duration )
Low
( k* #tasks +)
20
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Conclusion & Future Work
• Ignition Model
• Posting HIT: Cheap, but Slow
• Retainer: Fast, but Costly
-> Ignition balances both
• 10-Month-Long Deployment Study
• Adding a light-weight retainer speeds up recruiting
• Deploying a low-latency crowd-powered system is feasible
• What’s Next?
• Improving Ignition to address workers’ feedback
• Code Release Soon!
21
[ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ]
Thank you!
@windx0303 KennethHuang.cc

More Related Content

Similar to A 10-Month-Long Deployment Study of On-Demand Recruiting for Low-Latency Crowdsourcing

Making your analytics talk business | Big Data Demystified
Making your analytics talk business | Big Data DemystifiedMaking your analytics talk business | Big Data Demystified
Making your analytics talk business | Big Data DemystifiedOmid Vahdaty
 
Amreek dmaic template pph_may 14 project
Amreek dmaic template pph_may 14 projectAmreek dmaic template pph_may 14 project
Amreek dmaic template pph_may 14 projectamreek singh
 
Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013
Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013
Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013Kevin Halter
 
Use Agile to Supercharge your Job Search
Use Agile to Supercharge your Job SearchUse Agile to Supercharge your Job Search
Use Agile to Supercharge your Job SearchAlex McCarthy
 
Six Sigma Green Belt Project_GautamSingh
Six Sigma Green Belt Project_GautamSinghSix Sigma Green Belt Project_GautamSingh
Six Sigma Green Belt Project_GautamSinghGautam Singh
 
Have you tried to lean your IT Service Management? by Daniel Breston
Have you tried to lean your IT Service Management? by Daniel BrestonHave you tried to lean your IT Service Management? by Daniel Breston
Have you tried to lean your IT Service Management? by Daniel BrestonInstitut Lean France
 
Talent Analytics ERE 2015
Talent Analytics ERE 2015 Talent Analytics ERE 2015
Talent Analytics ERE 2015 Rob McIntosh
 
[Ania} Kanban Brazil In the depths of Kanban.pdf
[Ania} Kanban Brazil In the depths of Kanban.pdf[Ania} Kanban Brazil In the depths of Kanban.pdf
[Ania} Kanban Brazil In the depths of Kanban.pdfAnnaRadzikowskaACCA
 
AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...
AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...
AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...Amazon Web Services
 
SAP TechEd EXP17583 expert session SAP Screen Personas
SAP TechEd EXP17583 expert session SAP Screen PersonasSAP TechEd EXP17583 expert session SAP Screen Personas
SAP TechEd EXP17583 expert session SAP Screen PersonasPeter Spielvogel
 
Enhancement Training Course Program Curriculum
Enhancement Training Course Program CurriculumEnhancement Training Course Program Curriculum
Enhancement Training Course Program CurriculumJason Tenebro
 
Value Stream Mapping
Value Stream MappingValue Stream Mapping
Value Stream MappingTKMG, Inc.
 
APRA_Contact Reports_2016_Turner_Hrubik_IJM
APRA_Contact Reports_2016_Turner_Hrubik_IJMAPRA_Contact Reports_2016_Turner_Hrubik_IJM
APRA_Contact Reports_2016_Turner_Hrubik_IJMThomas Turner
 
OnRecruit-RPO PowerPoint Presentation
OnRecruit-RPO PowerPoint PresentationOnRecruit-RPO PowerPoint Presentation
OnRecruit-RPO PowerPoint PresentationChris Kitterman
 
Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...
Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...
Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...Institut Lean France
 
Allstate-T&M for ITSM-Kirch Final ipad
Allstate-T&M for ITSM-Kirch Final ipadAllstate-T&M for ITSM-Kirch Final ipad
Allstate-T&M for ITSM-Kirch Final ipadCathy Kirch
 

Similar to A 10-Month-Long Deployment Study of On-Demand Recruiting for Low-Latency Crowdsourcing (20)

Making your analytics talk business | Big Data Demystified
Making your analytics talk business | Big Data DemystifiedMaking your analytics talk business | Big Data Demystified
Making your analytics talk business | Big Data Demystified
 
Amreek dmaic template pph_may 14 project
Amreek dmaic template pph_may 14 projectAmreek dmaic template pph_may 14 project
Amreek dmaic template pph_may 14 project
 
Process analysis
Process analysisProcess analysis
Process analysis
 
EtaMax services
EtaMax servicesEtaMax services
EtaMax services
 
Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013
Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013
Top Mobile Apps for Construction Job-Sites_AGC Fall Conference 2013
 
Use Agile to Supercharge your Job Search
Use Agile to Supercharge your Job SearchUse Agile to Supercharge your Job Search
Use Agile to Supercharge your Job Search
 
Six Sigma Green Belt Project_GautamSingh
Six Sigma Green Belt Project_GautamSinghSix Sigma Green Belt Project_GautamSingh
Six Sigma Green Belt Project_GautamSingh
 
Have you tried to lean your IT Service Management? by Daniel Breston
Have you tried to lean your IT Service Management? by Daniel BrestonHave you tried to lean your IT Service Management? by Daniel Breston
Have you tried to lean your IT Service Management? by Daniel Breston
 
Talent Analytics ERE 2015
Talent Analytics ERE 2015 Talent Analytics ERE 2015
Talent Analytics ERE 2015
 
Smart Jacket - corrected one.
Smart Jacket  - corrected one.Smart Jacket  - corrected one.
Smart Jacket - corrected one.
 
[Ania} Kanban Brazil In the depths of Kanban.pdf
[Ania} Kanban Brazil In the depths of Kanban.pdf[Ania} Kanban Brazil In the depths of Kanban.pdf
[Ania} Kanban Brazil In the depths of Kanban.pdf
 
AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...
AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...
AWS re:Invent 2016: Leverage the Power of the Crowd To Work with Amazon Mecha...
 
SAP TechEd EXP17583 expert session SAP Screen Personas
SAP TechEd EXP17583 expert session SAP Screen PersonasSAP TechEd EXP17583 expert session SAP Screen Personas
SAP TechEd EXP17583 expert session SAP Screen Personas
 
Enhancement Training Course Program Curriculum
Enhancement Training Course Program CurriculumEnhancement Training Course Program Curriculum
Enhancement Training Course Program Curriculum
 
Value Stream Mapping
Value Stream MappingValue Stream Mapping
Value Stream Mapping
 
APRA_Contact Reports_2016_Turner_Hrubik_IJM
APRA_Contact Reports_2016_Turner_Hrubik_IJMAPRA_Contact Reports_2016_Turner_Hrubik_IJM
APRA_Contact Reports_2016_Turner_Hrubik_IJM
 
every-second-counts_QP_feb2016
every-second-counts_QP_feb2016every-second-counts_QP_feb2016
every-second-counts_QP_feb2016
 
OnRecruit-RPO PowerPoint Presentation
OnRecruit-RPO PowerPoint PresentationOnRecruit-RPO PowerPoint Presentation
OnRecruit-RPO PowerPoint Presentation
 
Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...
Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...
Pulling the flow in IT, what to expect? by Emmanuel Richard, BNP Paribas at t...
 
Allstate-T&M for ITSM-Kirch Final ipad
Allstate-T&M for ITSM-Kirch Final ipadAllstate-T&M for ITSM-Kirch Final ipad
Allstate-T&M for ITSM-Kirch Final ipad
 

More from Ting-Hao Huang

A Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over TimeA Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over TimeTing-Hao Huang
 
Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...
Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...
Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...Ting-Hao Huang
 
Real-time On-Demand Crowd-powered Entity Extraction
Real-time On-Demand Crowd-powered Entity ExtractionReal-time On-Demand Crowd-powered Entity Extraction
Real-time On-Demand Crowd-powered Entity ExtractionTing-Hao Huang
 
A Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over TimeA Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over TimeTing-Hao Huang
 
"Is there anything else I can help you with?": Challenges in Deploying an On-...
"Is there anything else I can help you with?": Challenges in Deploying an On-..."Is there anything else I can help you with?": Challenges in Deploying an On-...
"Is there anything else I can help you with?": Challenges in Deploying an On-...Ting-Hao Huang
 
Visual Storytelling (NAACL 2016, Poster)
Visual Storytelling (NAACL 2016, Poster)Visual Storytelling (NAACL 2016, Poster)
Visual Storytelling (NAACL 2016, Poster)Ting-Hao Huang
 
Social Metaphor Detection via Topical Analysis
Social Metaphor Detection via Topical AnalysisSocial Metaphor Detection via Topical Analysis
Social Metaphor Detection via Topical AnalysisTing-Hao Huang
 
Guardian: A Crowd-Powered Spoken Dialog System for Web APIs
Guardian: A Crowd-Powered Spoken Dialog System for Web APIsGuardian: A Crowd-Powered Spoken Dialog System for Web APIs
Guardian: A Crowd-Powered Spoken Dialog System for Web APIsTing-Hao Huang
 

More from Ting-Hao Huang (8)

A Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over TimeA Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over Time
 
Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...
Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...
Evorus: A Crowd-Powered Conversational Assistant Built to Automate Itself Ove...
 
Real-time On-Demand Crowd-powered Entity Extraction
Real-time On-Demand Crowd-powered Entity ExtractionReal-time On-Demand Crowd-powered Entity Extraction
Real-time On-Demand Crowd-powered Entity Extraction
 
A Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over TimeA Crowd-Powered Conversational Assistant That Automates Itself Over Time
A Crowd-Powered Conversational Assistant That Automates Itself Over Time
 
"Is there anything else I can help you with?": Challenges in Deploying an On-...
"Is there anything else I can help you with?": Challenges in Deploying an On-..."Is there anything else I can help you with?": Challenges in Deploying an On-...
"Is there anything else I can help you with?": Challenges in Deploying an On-...
 
Visual Storytelling (NAACL 2016, Poster)
Visual Storytelling (NAACL 2016, Poster)Visual Storytelling (NAACL 2016, Poster)
Visual Storytelling (NAACL 2016, Poster)
 
Social Metaphor Detection via Topical Analysis
Social Metaphor Detection via Topical AnalysisSocial Metaphor Detection via Topical Analysis
Social Metaphor Detection via Topical Analysis
 
Guardian: A Crowd-Powered Spoken Dialog System for Web APIs
Guardian: A Crowd-Powered Spoken Dialog System for Web APIsGuardian: A Crowd-Powered Spoken Dialog System for Web APIs
Guardian: A Crowd-Powered Spoken Dialog System for Web APIs
 

Recently uploaded

2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...Martijn de Jong
 
08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking Men08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking MenDelhi Call girls
 
Scaling API-first – The story of a global engineering organization
Scaling API-first – The story of a global engineering organizationScaling API-first – The story of a global engineering organization
Scaling API-first – The story of a global engineering organizationRadu Cotescu
 
Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)wesley chun
 
08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking MenDelhi Call girls
 
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdfThe Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdfEnterprise Knowledge
 
The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024Rafal Los
 
GenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day PresentationGenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day PresentationMichael W. Hawkins
 
The Codex of Business Writing Software for Real-World Solutions 2.pptx
The Codex of Business Writing Software for Real-World Solutions 2.pptxThe Codex of Business Writing Software for Real-World Solutions 2.pptx
The Codex of Business Writing Software for Real-World Solutions 2.pptxMalak Abu Hammad
 
Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024The Digital Insurer
 
Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsJoaquim Jorge
 
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...apidays
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfsudhanshuwaghmare1
 
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...Neo4j
 
A Call to Action for Generative AI in 2024
A Call to Action for Generative AI in 2024A Call to Action for Generative AI in 2024
A Call to Action for Generative AI in 2024Results
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationSafe Software
 
Breaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path MountBreaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path MountPuma Security, LLC
 
Presentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreterPresentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreternaman860154
 
Histor y of HAM Radio presentation slide
Histor y of HAM Radio presentation slideHistor y of HAM Radio presentation slide
Histor y of HAM Radio presentation slidevu2urc
 
🐬 The future of MySQL is Postgres 🐘
🐬  The future of MySQL is Postgres   🐘🐬  The future of MySQL is Postgres   🐘
🐬 The future of MySQL is Postgres 🐘RTylerCroy
 

Recently uploaded (20)

2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...
 
08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking Men08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking Men
 
Scaling API-first – The story of a global engineering organization
Scaling API-first – The story of a global engineering organizationScaling API-first – The story of a global engineering organization
Scaling API-first – The story of a global engineering organization
 
Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)
 
08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men08448380779 Call Girls In Friends Colony Women Seeking Men
08448380779 Call Girls In Friends Colony Women Seeking Men
 
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdfThe Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
 
The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024
 
GenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day PresentationGenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day Presentation
 
The Codex of Business Writing Software for Real-World Solutions 2.pptx
The Codex of Business Writing Software for Real-World Solutions 2.pptxThe Codex of Business Writing Software for Real-World Solutions 2.pptx
The Codex of Business Writing Software for Real-World Solutions 2.pptx
 
Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024
 
Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and Myths
 
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdf
 
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
 
A Call to Action for Generative AI in 2024
A Call to Action for Generative AI in 2024A Call to Action for Generative AI in 2024
A Call to Action for Generative AI in 2024
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
 
Breaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path MountBreaking the Kubernetes Kill Chain: Host Path Mount
Breaking the Kubernetes Kill Chain: Host Path Mount
 
Presentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreterPresentation on how to chat with PDF using ChatGPT code interpreter
Presentation on how to chat with PDF using ChatGPT code interpreter
 
Histor y of HAM Radio presentation slide
Histor y of HAM Radio presentation slideHistor y of HAM Radio presentation slide
Histor y of HAM Radio presentation slide
 
🐬 The future of MySQL is Postgres 🐘
🐬  The future of MySQL is Postgres   🐘🐬  The future of MySQL is Postgres   🐘
🐬 The future of MySQL is Postgres 🐘
 

A 10-Month-Long Deployment Study of On-Demand Recruiting for Low-Latency Crowdsourcing

  • 1. A 10-Month-Long Deployment Study of On-Demand Recruiting for Low-Latency Crowdsourcing Live Feedback & Questions: http://tinyurl.com/20171026IGNITION Ting-Hao (Kenneth) Huang Jeffrey P. Bigham Carnegie Mellon University Image: https://pixabay.com/photo-743920/
  • 2. 2 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Chorus: A Crowd-powered Conversation Assistant "Is there anything else I can help you with?": Challenges in Deploying an On-Demand Crowd-Powered Conversational Agent. Huang et. al., HCOMP 2016
  • 3. 3 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Constraints of Deploying Chorus • Small user traffic • Avg. 2.97 conversations per day • Fast worker response time • Response time to a message ~= 32 sec (student participant)* • Response time to a message ~= 105 sec (startup participant)* • Need multiple workers to work together • Limited budget • 24-7 service * IM waiting: timing and responsiveness in semi-synchronous communication. Avrahami et al. CSCW’08. …Similar to many academic projects !
  • 4. 4 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] How to recruit workers quickly on-demand? On-demand (QuikTurkit, Bigham, et al., ‘10) On-call (Retainer Model, Bernstein, et al.,‘11) Ignition Method Post HITs + SEO Retainer Hybrid Latency 60 - 120 seconds < 10 seconds 20 – 90 seconds How to Hold Workers Using old tasks Waiting Page Waiting Page Recruiting Cost Per Task Old tasks + SEO (mturk fee) All Workers’ Waiting Time 55% of Workers’ Waiting Time Recruiting Cost For Small Deployments Very Low ( k* #tasks ) High ( k* duration ) Low ( k* #tasks +) …Can we have something inexpensive and fast?
  • 5. 5 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Engines Generate Power On-demand (QuikTurkit, Bigham, et al., ‘10) On-call (Retainer Model, Bernstein, et al.,‘11) Ignition Method Post HITs + SEO Retainer Hybrid Latency 60 - 120 seconds < 10 seconds 20 – 90 seconds How to Hold Workers Using old tasks Waiting Page Waiting Page Recruiting Cost Per Task Old tasks + SEO (mturk fee) All Workers’ Waiting Time 55% of Workers’ Waiting Time Recruiting Cost For Small Deployments Very Low ( k* #tasks ) High ( k* duration ) Low ( k* #tasks +) • Crowd-powered System Why not both? Re-Start the Engine Each Time Start the Engine Once, and Hold the Power Image: https://commons.wikimedia.org/wiki/File:3791_cc_Holden_3800_V6_engine_(2015-08-29)_03.jpg
  • 6. 6 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Ignition Model 1. Hire slightly more workers than we need 2. Hold the extra workers in the retainer 3. Always call workers from retainer first, and then post HITs
  • 7. 7 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Ignition = Retainer + Market • When you want to recruit 8 workers each time… • A Chorus task need at most 5 workers #Workers Recruited from Retainer #Workers Recruited from Mturk Market ( Posting a HIT w/ Cmarket Assignments ) Posting More Assignments -> Gets Workers Faster Cretainer + Cmarket = 8
  • 8. 8 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Waiting Page & Point System http://TinyUrl.com/IgnitionDemo
  • 9. 9 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Transition Graph for Workers in Ignition
  • 10. 10 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Transition Graph for Workers in Ignition
  • 11. 11 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Transition Graph for Workers in Ignition
  • 12. 12 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] 1. The task occurs instantly [ Instant ] 2. The task occurs later [ Retained ] 3. No tasks at all [ No Task ] Transition Graph for Workers in Ignition
  • 13. 13 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Ignition Deployment • From July 1st, 2016 to April 27th, 2017 (~10 months) • 6,823 assignments, 648 workers • $0.25 for a 30-minute waiting time • Mturk Qualifications Used • HIT Approval Rate > 90%, #Approved HITs> 200, Adult Content Qualification Assignments Tasks (Conversation) #workers in retainer 50% of tasks occurred when 1+ workers in the retainer
  • 14. 14 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Posting More Assignments -> Gets Workers Faster Simply Posting HITs Getting 1+ Workers Getting 3+ Workers X-Axis: Time (x seconds) Y-Axis: Prob(Getting N+ workers within x seconds) Cmarket = 3 Cmarket = 8 Cmarket = 8 Cmarket = 3
  • 15. 15 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Simply Posting HITs v.s. Ignition Getting 1+ Workers Getting 3+ Workers Cmarket = 8 Cmarket = 8 Cmarket = 8 Cmarket = 8
  • 16. 16 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Does a Deployed Retainer Work? • Yes, and (surprisingly) stable. • Response rate = Proportion of workers in the retainer who responded within 20 secs when being called Response rate is stable over time. Month of a Year Day of a Week Retainer Time
  • 17. 17 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] What do workers think? • Worker Survey • Collect opinions and self-reported behavior • Ask 156 workers who completed 10+ Ignition HITs • $2 compensation (~10mins) • 101 workers finished the survey (Response Rate = 64.74%)
  • 18. 18 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] General HITs v.s. Ignition HITs "I like doing HITs on MTurk in general." "I like doing Retainer HITs." • 5-point Likert scale • Strongly Disagree to Strongly Agree
  • 19. 19 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] How to recruit workers quickly on-demand? On-demand (QuikTurkit, Bigham, et al., ‘10) On-call (Retainer Model, Bernstein, et al.,‘11) Ignition Method Post HITs + SEO Retainer Hybrid Latency 60 - 120 seconds < 10 seconds 20 – 90 seconds How to Hold Workers Using old tasks Waiting Page Waiting Page Recruiting Cost Per Task Old tasks + SEO (mturk fee) All Workers’ Waiting Time 55% of Workers’ Waiting Time Recruiting Cost For Small Deployment Very Low ( k* #tasks ) High ( k* duration ) Low ( k* #tasks +)
  • 20. 20 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Conclusion & Future Work • Ignition Model • Posting HIT: Cheap, but Slow • Retainer: Fast, but Costly -> Ignition balances both • 10-Month-Long Deployment Study • Adding a light-weight retainer speeds up recruiting • Deploying a low-latency crowd-powered system is feasible • What’s Next? • Improving Ignition to address workers’ feedback • Code Release Soon!
  • 21. 21 [ Live Feedback & Questions: http://tinyurl.com/20171026Ignition ] Thank you! @windx0303 KennethHuang.cc