Crowdsourcing For Research and Engineering (Tutorial given at CrowdConf 2011)

Crowdsourcing For Research and Engineering (Tutorial given at CrowdConf 2011)
Crowdsourcing for Research & Engineering Omar Alonso Microsoft Matthew Lease University of Texas at Austin November 1, 2011 November 1, 2011 Crowdsourcing for Research and Engineering 1
Tutorial Objectives • What is crowdsourcing? • How and when to use crowdsourcing? • How to use Mechanical Turk • Experimental setup and design guidelines for working with the crowd • Quality control: issues, measuring, and improving • Future trends • Research landscape and open challenges November 1, 2011 Crowdsourcing for Research and Engineering 2
Tutorial Outline I. Introduction to Crowdsourcing I. Introduction, Examples, Terminology II. Primary focus on micro-tasks II. Tools and Platforms: APIs and Examples III. Methodology for effective crowdsourcing I. Methods, Examples, and Tips II. Quality control: monitoring & improving IV. Future Trends November 1, 2011 Crowdsourcing for Research and Engineering 3
I INTRODUCTION TO CROWDSOURCING November 1, 2011 Crowdsourcing for Research and Engineering 4
From Outsourcing to Crowdsourcing • Take a job traditionally performed by a known agent (often an employee) • Outsource it to an undefined, generally large group of people via an open call • New application of principles from open source movement • Evolving & broadly defined ... November 1, 2011 Crowdsourcing for Research and Engineering 5
Examples Crowdsourcing 101: Putting the WSDM of Crowds to Work for You. 6
November 1, 2011 Crowdsourcing for Research and Engineering 7
Crowdsourcing models • Virtual work, Micro-tasks, & Aggregators • Open Innovation, Co-Creation, & Contests • Citizen Science • Prediction Markets • Crowd Funding and Charity • “Gamification” (not serious gaming) • Transparent • cQ&A, Social Search, and Polling • Human Sensing November 1, 2011 Crowdsourcing for Research and Engineering 8
What is Crowdsourcing? • A collection of mechanisms and associated methodologies for scaling and directing crowd activities to achieve some goal(s) • Enabled by internet-connectivity • Many related areas – Collective intelligence – Social computing – People services – Human computation (next slide…) • Good work is creative, innovative, surprising, … November 1, 2011 Crowdsourcing for Research and Engineering 9
Human Computation • Having people do stuff instead of computers • Investigates use of people to execute certain computations for which capabilities of current automated methods are more limited • Explores the metaphor of computation for characterizing attributes, capabilities, and limitations of human performance in executing desired tasks • Computation is required, crowd is not • Pioneer: Luis von Ahn’s thesis (2005) November 1, 2011 Crowdsourcing for Research and Engineering 10
What is not crowdsourcing? • Ingredients necessary but not sufficient – A crowd – Digital communication • Post-hoc use of undirected crowd behaviors – e.g. Data mining, visualization • Conducting a traditional survey or poll • Human Computation with one or few people – E.g. traditional active learning • … November 1, 2011 Crowdsourcing for Research and Engineering 11
Crowdsourcing Key Questions • What are the goals? – Purposeful directing of human activity • How can you incentivize participation? – Incentive engineering – Who are the target participants? • Which model(s) are most appropriate? – How to adapt them to your context and goals? November 1, 2011 Crowdsourcing for Research and Engineering 12
What do you want to accomplish? • Perform specified task(s) • Innovate and/or discover • Create • Predict • Fund • Learn • Monitor November 1, 2011 Crowdsourcing for Research and Engineering
Why Should Anyone Participate? Don’t let this happen to you … November 1, 2011 Crowdsourcing for Research and Engineering 14
Incentive Engineering • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige (leaderboards, badges) • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource Multiple incentives can often operate in parallel (*caveat) November 1, 2011 Crowdsourcing for Research and Engineering 15
Models: Goal(s) + Incentives • Virtual work, Micro-tasks, & Aggregators • Open Innovation, Co-Creation, & Contests • Citizen Science • Prediction Markets • Crowd Funding and Charity • “Gamification” (not serious gaming) • Transparent • cQ&A, Social Search, and Polling • Human Sensing November 1, 2011 Crowdsourcing for Research and Engineering 16
Example: Wikipedia • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 17
Example: • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 18
Example: ESP and GWaP L. Von Ahn and L. Dabbish (2004) November 1, 2011 Crowdsourcing for Research and Engineering 19
Example: ESP • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 20
Example: fold.it S. Cooper et al. (2010) Alice G. Walton. Online Gamers Help Solve Mystery of Critical AIDS Virus Enzyme. The Atlantic, October 8, 2011. November 1, 2011 Crowdsourcing for Research and Engineering 21
Example: fold.it • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 22
Example: FreeRice November 1, 2011 Crowdsourcing for Research and Engineering 23
Example: FreeRice • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 24
Example: cQ&A, Social Search, & Polling November 1, 2011 Crowdsourcing for Research and Engineering 25
Example: cQ&A • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 26
Example: reCaptcha November 1, 2011 Crowdsourcing for Research and Engineering 27
Example: reCaptcha • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) Is there an existing human activity you can harness • Learn something new for another purpose? • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 28
Example: Mechanical Turk J. Pontin. Artificial Intelligence, With Help From the Humans. New York Times (March 25, 2007) November 1, 2011 Crowdsourcing for Research and Engineering 29
Example: Mechanical Turk • Earn Money (real or virtual) • Have fun (or pass the time) • Socialize with others • Obtain recognition or prestige • Do Good (altruism) • Learn something new • Obtain something else • Create self-serving resource November 1, 2011 Crowdsourcing for Research and Engineering 30
Look Before You Leap a • Wolfson & Lease (2011) • Identify a few potential legal pitfalls to know about when considering crowdsourcing – employment law – patent inventorship – data security and the Federal Trade Commission – copyright ownership – securities regulation of crowdfunding • Take-away: don’t panic, just be mindful of the law November 1, 2011 Crowdsourcing for Research and Engineering 31
Example: SamaSource Incentive for YOU: Do Good Terminology: channels November 1, 2011 Crowdsourcing for Research and Engineering 32
Who are the workers? • A. Baio, November 2008. The Faces of Mechanical Turk. • P. Ipeirotis. March 2010. The New Demographics of Mechanical Turk • J. Ross, et al. Who are the Crowdworkers?... CHI 2010. November 1, 2011 Crowdsourcing for Research and Engineering 33
MTurk Demographics • 2008-2009 studies found less global and diverse than previously thought – US – Female – Educated – Bored – Money is secondary November 1, 2011 Crowdsourcing for Research and Engineering 34
2010 shows increasing diversity 47% US, 34% India, 19% other (P. Ipeitorotis. March 2010) November 1, 2011 Crowdsourcing for Research and Engineering 35
MICRO-TASKS++ AND EXAMPLES November 1, 2011 Crowdsourcing for Research and Engineering 36
Chess machine unveiled in 1770 by Wolfgang von Kempelen (1734–1804) • “Micro-task” crowdsourcing marketplace • On-demand, scalable, real-time workforce • Online since 2005 (and still in “beta”) • Programmer’s API & “Dashboard” GUI • Sponsorship: TREC 2011 Crowdsourcing Track (pending) November 1, 2011 Crowdsourcing for Research and Engineering 37
Does anyone really use it? Yes! http://www.mturk-tracker.com (P. Ipeirotis’10) From 1/09 – 4/10, 7M HITs from 10K requestors worth $500,000 USD (significant under-estimate) November 1, 2011 Crowdsourcing for Research and Engineering 38
• Labor on-demand, Channels, Quality control features • Sponsorship – Research Workshops: CSE’10, CSDM’11, CIR’11, – TREC 2011 Crowdsourcing Track November 1, 2011 Crowdsourcing for Research and Engineering 39
CloudFactory • Information below from Mark Sears (Oct. 18, 2011) • Cloud Labor API – Tools to design virtual assembly lines – workflows with multiple tasks chained together • Focus on self serve tools for people to easily design crowd-powered assembly lines that can be easily integrated into software applications • Interfaces: command-line, RESTful API, and Web • Each “task station” can have either a human or robot worker assigned – web software services (AlchemyAPI, SendGrid, Google APIs, Twilio, etc.) or local software can be combined with human computation • Many built-in "best practices" – “Tournament Stations” where multiple results are compared by a other cloud workers until confidence of best answer is reached – “Improver Stations” have workers improve and correct work by other workers – Badges are earned by cloud workers passing tests created by requesters – Training and tools to create skill tests will be flexible – Algorithms to detect and kick out spammers/cheaters/lazy/bad workers • Sponsorship: TREC 2012 Crowdsourcing Track November 1, 2011 Crowdsourcing for Research and Engineering 40
More Crowd Labor Platforms • Clickworker • CloudCrowd • CrowdSource • DoMyStuff • Humanoid (by Matt Swason et al.) • Microtask • MobileWorks (by Anand Kulkarni ) • myGengo • SmartSheet • vWorker • Industry heavy-weights – Elance – Liveops – oDesk – uTest • and more… November 1, 2011 Crowdsourcing for Research and Engineering 41
Why Micro-Tasks? • Easy, cheap and fast • Ready-to use infrastructure, e.g. – MTurk payments, workforce, interface widgets – CrowdFlower quality control mechanisms, etc. – Many others … • Allows early, iterative, frequent trials – Iteratively prototype and test new ideas – Try new tasks, test when you want & as you go • Many successful examples of use reported November 1, 2011 Crowdsourcing for Research and Engineering 42
Micro-Task Issues • Process – Task design, instructions, setup, iteration • Choose crowdsourcing platform (or roll your own) • Human factors – Payment / incentives, interface and interaction design, communication, reputation, recruitment, retention • Quality Control / Data Quality – Trust, reliability, spam detection, consensus labeling November 1, 2011 Crowdsourcing for Research and Engineering 43
Legal Disclaimer: Caution Tape and Silver Bullets • Often still involves more art than science • Not a magic panacea, but another alternative – one more data point for analysis, complements other methods • Quality may be traded off for time/cost/effort • Hard work & experimental design still required! November 1, 2011 Crowdsourcing for Research and Engineering 44
Hello World Demo • We’ll show a simple, short demo of MTurk • This is a teaser highlighting things we’ll discuss – Don’t worry about details; we’ll revisit them • Specific task unimportant • Big idea: easy, fast, cheap to label with MTurk! November 1, 2011 Crowdsourcing for Research and Engineering 45
Jane saw the man with the binoculars November 1, 2011 Crowdsourcing for Research and Engineering 46
DEMO November 1, 2011 Crowdsourcing for Research and Engineering 47
Traditional Data Collection • Setup data collection software / harness • Recruit participants • Pay a flat fee for experiment or hourly wage • Characteristics – Slow – Expensive – Tedious – Sample Bias November 1, 2011 Crowdsourcing for Research and Engineering 48
Research Using Micro-Tasks • Let’s see examples of micro-task usage – Many areas: IR, NLP, computer vision, user studies, usability testing, psychological studies, surveys, … • Check bibliography at end for more references November 1, 2011 Crowdsourcing for Research and Engineering 49
NLP Example – Dialect Identification November 1, 2011 Crowdsourcing for Research and Engineering 50
NLP Example – Spelling correction November 1, 2011 Crowdsourcing for Research and Engineering 51
NLP Example – Machine Translation • Manual evaluation on translation quality is slow and expensive • High agreement between non-experts and experts • $0.10 to translate a sentence C. Callison-Burch. “Fast, Cheap, and Creative: Evaluating Translation Quality Using Amazon’s Mechanical Turk”, EMNLP 2009. B. Bederson et al. Translation by Interactive Collaboration between Monolingual Users, GI 2010 November 1, 2011 Crowdsourcing for Research and Engineering 52
NLP Example – Snow et al. (2008) • 5 Tasks – Affect recognition – Word similarity – Recognizing textual entailment – Event temporal ordering – Word sense disambiguation • high agreement between crowd labels and expert “gold” labels – assumes training data for worker bias correction • 22K labels for $26 ! November 1, 2011 Crowdsourcing for Research and Engineering 53
Computer Vision – Painting Similarity Kovashka & Lease, CrowdConf’10 November 1, 2011 Crowdsourcing for Research and Engineering 54
User Studies • Investigate attitudes about saving, sharing, publishing, and removing online photos • Survey – A scenario-based probe of respondent attitudes, designed to yield quantitative data – A set of questions (close and open-ended) – Importance of recent activity – 41 question – 7 point scale • 250 respondents C. Marshall and F. Shipman. “The Ownership and Reuse of Visual Media”, JCDL 2011. November 1, 2011 Crowdsourcing for Research and Engineering 55
Remote Usability Testing • Liu et al. (in preparation) • Compares remote usability testing using MTurk and CrowdFlower (not uTest) vs. traditional on-site testing • Advantages – More Participants – More Diverse Participants – High Speed – Low Cost • Disadvantages – Lower Quality Feedback – Less Interaction – Greater need for quality control – Less Focused User Groups November 1, 2011 Crowdsourcing for Research and Engineering 56
IR Example – Relevance and ads November 1, 2011 Crowdsourcing for Research and Engineering 57
IR Example – Product Search November 1, 2011 Crowdsourcing for Research and Engineering 58
IR Example – Snippet Evaluation • Study on summary lengths • Determine preferred result length • Asked workers to categorize web queries • Asked workers to evaluate snippet quality • Payment between $0.01 and $0.05 per HIT M. Kaisser, M. Hearst, and L. Lowe. “Improving Search Results Quality by Customizing Summary Lengths”, ACL/HLT, 2008. November 1, 2011 Crowdsourcing for Research and Engineering 59
IR Example – Relevance Assessment • Replace TREC-like relevance assessors with MTurk? • Selected topic “space program” (011) • Modified original 4-page instructions from TREC • Workers more accurate than original assessors! • 40% provided justification for each answer O. Alonso and S. Mizzaro. “Can we get rid of TREC assessors? Using Mechanical Turk for relevance assessment”, SIGIR Workshop on the Future of IR Evaluation, 2009. November 1, 2011 Crowdsourcing for Research and Engineering 60
IR Example – Timeline Annotation • Workers annotate timeline on politics, sports, culture • Given a timex (1970s, 1982, etc.) suggest something • Given an event (Vietnam, World cup, etc.) suggest a timex K. Berberich, S. Bedathur, O. Alonso, G. Weikum “A Language Modeling Approach for Temporal Information Needs”. ECIR 2010 November 1, 2011 Crowdsourcing for Research and Engineering 61
How can I get started? • You have an idea • Easy, cheap, fast, and iterative sounds good Can you test your idea via crowdsourcing? • Is my idea crowdsourcable? • How do I start? • What do I need? November 1, 2011 Crowdsourcing for Research and Engineering 62
Tip for Getting Started: do work Try doing work before you create work for others! November 1, 2011 Crowdsourcing for Research and Engineering 63
II AMAZON MECHANICAL TURK November 1, 2011 Crowdsourcing for Research and Engineering 64
Mechanical What? November 1, 2011 Crowdsourcing for Research and Engineering 65
MTurk: The Requester • Sign up with your Amazon account • Amazon payments • Purchase prepaid HITs • There is no minimum or up-front fee • MTurk collects a 10% commission • The minimum commission charge is $0.005 per HIT November 1, 2011 Crowdsourcing for Research and Engineering 66
MTurk Dashboard • Three tabs – Design – Publish – Manage • Design – HIT Template • Publish – Make work available • Manage – Monitor progress November 1, 2011 Crowdsourcing for Research and Engineering 67
MTurk: Dashboard - II November 1, 2011 Crowdsourcing for Research and Engineering 68
MTurk API • Amazon Web Services API • Rich set of services • Command line tools • More flexibility than dashboard November 1, 2011 Crowdsourcing for Research and Engineering 69
MTurk Dashboard vs. API • Dashboard – Easy to prototype – Setup and launch an experiment in a few minutes • API – Ability to integrate AMT as part of a system – Ideal if you want to run experiments regularly – Schedule tasks November 1, 2011 Crowdsourcing for Research and Engineering 70
Working on MTurk • Sign up with your Amazon account • Tabs – Account: work approved/rejected – HIT: browse and search for work – Qualifications: browse & search qualifications • Start turking! November 1, 2011 Crowdsourcing for Research and Engineering 71
Why Eytan Adar hates MTurk Research (at least sort of) • Overly-narrow focus on Turk & other platforms – Identify general vs. platform-specific problems – Academic vs. Industrial problems • Lack of appreciation of interdisciplinary nature – Some problems well-studied in other areas – Human behavior hasn’t changed much • Turks aren’t Martians – How many prior user studies do we have to reproduce on MTurk before we can get over it? November 1, 2011 Crowdsourcing for Research and Engineering 72
III RELEVANCE JUDGING & CROWDSOURCING November 1, 2011 Crowdsourcing for Research and Engineering 73
Motivating Example: Relevance Judging • Relevance of search results is difficult to judge – Highly subjective – Expensive to measure • Professional editors commonly used • Potential benefits of crowdsourcing – Scalability (time and cost) – Diversity of judgments November 1, 2011 Crowdsourcing for Research and Engineering 75
Started with a joke … November 1, 2011 Crowdsourcing for Research and Engineering 77
Results for {idiot} at WSDM 2011 February 2011: 5/7 (R), 2/7 (NR) – Most of the time those TV reality stars have absolutely no talent. They do whatever they can to make a quick dollar. Most of the time the reality tv stars don not have a mind of their own. R – Most are just celebrity wannabees. Many have little or no talent, they just want fame. R – I can see this one going both ways. A particular sort of reality star comes to mind, though, one who was voted off Survivor because he chose not to use his immunity necklace. Sometimes the label fits, but sometimes it might be unfair. R – Just because someone else thinks they are an "idiot", doesn't mean that is what the word means. I don't like to think that any one person's photo would be used to describe a certain term. NR – While some reality-television stars are genuinely stupid (or cultivate an image of stupidity), that does not mean they can or should be classified as "idiots." Some simply act that way to increase their TV exposure and potential earnings. Other reality-television stars are really intelligent people, and may be considered as idiots by people who don't like them or agree with them. It is too subjective an issue to be a good result for a search engine. NR – Have you seen the knuckledraggers on reality television? They should be required to change their names to idiot after appearing on the show. You could put numbers after the word idiot so we can tell them apart. R – Although I have not followed too many of these shows, those that I have encountered have for a great part a very common property. That property is that most of the participants involved exhibit a shallow self-serving personality that borders on social pathological behavior. To perform or act in such an abysmal way could only be an act of an idiot. R November 1, 2011 Crowdsourcing for Research and Engineering 78
Two Simple Examples of MTurk 1. Ask workers to classify a query 2. Ask workers to judge document relevance Steps • Define high-level task • Design & implement interface & backend • Launch, monitor progress, and assess work • Iterate design November 1, 2011 Crowdsourcing for Research and Engineering 79
Query Classification Task • Ask the user to classify a query • Show a form that contains a few categories • Upload a few queries (~20) • Use 3 workers November 1, 2011 Crowdsourcing for Research and Engineering 80
Relevance Judging Task • Use a few documents from a standard collection used for evaluating search engines • Ask user to make binary judgments • Modification: graded judging • Use 5 workers November 1, 2011 Crowdsourcing for Research and Engineering 83
IV METHODOLOGY FOR EFFECTIVE CROWDSOURCING November 1, 2011 Crowdsourcing for Research and Engineering 85
Typical Workflow • Define and design what to test • Sample data • Design the experiment • Run experiment • Collect data and analyze results • Quality control November 1, 2011 Crowdsourcing for Research and Engineering 87
Development Framework • Incremental approach • Measure, evaluate, and adjust as you go • Suitable for repeatable tasks November 1, 2011 Crowdsourcing for Research and Engineering 88
Survey Design • One of the most important parts • Part art, part science • Instructions are key • Prepare to iterate November 1, 2011 Crowdsourcing for Research and Engineering 89
Questionnaire Design • Ask the right questions • Workers may not be IR experts so don’t assume the same understanding in terms of terminology • Show examples • Hire a technical writer – Engineer writes the specification – Writer communicates November 1, 2011 Crowdsourcing for Research and Engineering 90
UX Design • Time to apply all those usability concepts • Generic tips – Experiment should be self-contained. – Keep it short and simple. Brief and concise. – Be very clear with the relevance task. – Engage with the worker. Avoid boring stuff. – Always ask for feedback (open-ended question) in an input box. November 1, 2011 Crowdsourcing for Research and Engineering 91
UX Design - II • Presentation • Document design • Highlight important concepts • Colors and fonts • Need to grab attention • Localization November 1, 2011 Crowdsourcing for Research and Engineering 92
Examples - I • Asking too much, task not clear, “do NOT/reject” • Worker has to do a lot of stuff November 1, 2011 Crowdsourcing for Research and Engineering 93
Example - II • Lot of work for a few cents • Go here, go there, copy, enter, count … November 1, 2011 Crowdsourcing for Research and Engineering 94
A Better Example • All information is available – What to do – Search result – Question to answer November 1, 2011 Crowdsourcing for Research and Engineering 95
Form and Metadata • Form with a close question (binary relevance) and open-ended question (user feedback) • Clear title, useful keywords • Workers need to find your task November 1, 2011 Crowdsourcing for Research and Engineering 97
Relevance Judging – Example I November 1, 2011 Crowdsourcing for Research and Engineering 98
Relevance Judging – Example II November 1, 2011 Crowdsourcing for Research and Engineering 99
Implementation • Similar to a UX • Build a mock up and test it with your team – Yes, you need to judge some tasks • Incorporate feedback and run a test on MTurk with a very small data set – Time the experiment – Do people understand the task? • Analyze results – Look for spammers – Check completion times • Iterate and modify accordingly November 1, 2011 Crowdsourcing for Research and Engineering 100
Implementation – II • Introduce quality control – Qualification test – Gold answers (honey pots) • Adjust passing grade and worker approval rate • Run experiment with new settings & same data • Scale on data • Scale on workers November 1, 2011 Crowdsourcing for Research and Engineering 101
Experiment in Production • Lots of tasks on MTurk at any moment • Need to grab attention • Importance of experiment metadata • When to schedule – Split a large task into batches and have 1 single batch in the system – Always review feedback from batch n before uploading n+1 November 1, 2011 Crowdsourcing for Research and Engineering 102
How Much to Pay? • Price commensurate with task effort – Ex: $0.02 for yes/no answer + $0.02 bonus for optional feedback • Ethics & market-factors: W. Mason and S. Suri, 2010. – e.g. non-profit SamaSource contracts workers refugee camps – Predict right price given market & task: Wang et al. CSDM’11 • Uptake & time-to-completion vs. Cost & Quality – Too little $$, no interest or slow – too much $$, attract spammers – Real problem is lack of reliable QA substrate • Accuracy & quantity – More pay = more work, not better (W. Mason and D. Watts, 2009) • Heuristics: start small, watch uptake and bargaining feedback • Worker retention (“anchoring”) See also: L.B. Chilton et al. KDD-HCOMP 2010. November 1, 2011 Crowdsourcing for Research and Engineering 103
Quality Control in General • Extremely important part of the experiment • Approach as “overall” quality; not just for workers • Bi-directional channel – You may think the worker is doing a bad job. – The same worker may think you are a lousy requester. November 1, 2011 Crowdsourcing for Research and Engineering 105
When to assess quality of work • Beforehand (prior to main task activity) – How: “qualification tests” or similar mechanism – Purpose: screening, selection, recruiting, training • During – How: assess labels as worker produces them • Like random checks on a manufacturing line – Purpose: calibrate, reward/penalize, weight • After – How: compute accuracy metrics post-hoc – Purpose: filter, calibrate, weight, retain (HR) – E.g. Jung & Lease (2011), Tang & Lease (2011), ... November 1, 2011 Crowdsourcing for Research and Engineering 106
How to assess quality of work? • Compare worker’s label vs. – Known (correct, trusted) label – Other workers’ labels • P. Ipeirotis. Worker Evaluation in Crowdsourcing: Gold Data or Multiple Workers? Sept. 2010. – Model predictions of the above • Model the labels (Ryu & Lease, ASIS&T11) • Model the workers (Chen et al., AAAI’10) • Verify worker’s label – Yourself – Tiered approach (e.g. Find-Fix-Verify) • Quinn and B. Bederson’09, Bernstein et al.’10 November 1, 2011 Crowdsourcing for Research and Engineering 107
Typical Assumptions • Objective truth exists – no minority voice / rare insights – Can relax this to model “truth distribution” • Automatic answer comparison/evaluation – What about free text responses? Hope from NLP… • Automatic essay scoring • Translation (BLEU: Papineni, ACL’2002) • Summarization (Rouge: C.Y. Lin, WAS’2004) – Have people do it (yourself or find-verify crowd, etc.) November 1, 2011 Crowdsourcing for Research and Engineering 108
Distinguishing Bias vs. Noise • Ipeirotis (HComp 2010) • People often have consistent, idiosyncratic skews in their labels (bias) – E.g. I like action movies, so they get higher ratings • Once detected, systematic bias can be calibrated for and corrected (yeah!) • Noise, however, seems random & inconsistent – this is the real issue we want to focus on November 1, 2011 Crowdsourcing for Research and Engineering 109
Comparing to known answers • AKA: gold, honey pot, verifiable answer, trap • Assumes you have known answers • Cost vs. Benefit – Producing known answers (experts?) – % of work spent re-producing them • Finer points – Controls against collusion – What if workers recognize the honey pots? November 1, 2011 Crowdsourcing for Research and Engineering 110
Comparing to other workers • AKA: consensus, plurality, redundant labeling • Well-known metrics for measuring agreement • Cost vs. Benefit: % of work that is redundant • Finer points – Is consensus “truth” or systematic bias of group? – What if no one really knows what they’re doing? • Low-agreement across workers indicates problem is with the task (or a specific example), not the workers – Risk of collusion • Sheng et al. (KDD 2008) November 1, 2011 Crowdsourcing for Research and Engineering 111
Comparing to predicted label • Ryu & Lease, ASIS&T11 (CrowdConf’11 poster) • Catch-22 extremes – If model is really bad, why bother comparing? – If model is really good, why collect human labels? • Exploit model confidence – Trust predictions proportional to confidence – What if model very confident and wrong? • Active learning – Time sensitive: Accuracy / confidence changes November 1, 2011 Crowdsourcing for Research and Engineering 112
Compare to predicted worker labels • Chen et al., AAAI’10 • Avoid inefficiency of redundant labeling – See also: Dekel & Shamir (COLT’2009) • Train a classifier for each worker • For each example labeled by a worker – Compare to predicted labels for all other workers • Issues • Sparsity: workers have to stick around to train model… • Time-sensitivity: New workers & incremental updates? November 1, 2011 Crowdsourcing for Research and Engineering 113
Methods for measuring agreement • What to look for – Agreement, reliability, validity • Inter-agreement level – Agreement between judges – Agreement between judges and the gold set • Some statistics – Percentage agreement – Cohen’s kappa (2 raters) – Fleiss’ kappa (any number of raters) – Krippendorff’s alpha • With majority vote, what if 2 say relevant, 3 say not? – Use expert to break ties (Kochhar et al, HCOMP’10; GQR) – Collect more judgments as needed to reduce uncertainty November 1, 2011 Crowdsourcing for Research and Engineering 114
Inter-rater reliability • Lots of research • Statistics books cover most of the material • Three categories based on the goals – Consensus estimates – Consistency estimates – Measurement estimates November 1, 2011 Crowdsourcing for Research and Engineering 115
Sample code – R packages psy and irr >library(psy) >library(irr) >my_data <- read.delim(file="test.txt", head=TRUE, sep="t") >kappam.fleiss(my_data,exact=FALSE) >my_data2 <- read.delim(file="test2.txt", head=TRUE, sep="t") >ckappa(my_data2) November 1, 2011 Crowdsourcing for Research and Engineering 116
k coefficient • Different interpretations of k • For practical purposes you need to be >= moderate • Results may vary k Interpretation <0 Poor agreement 0.01 – 0.20 Slight agreement 0.21 – 0.40 Fair agreement 0.41 – 0.60 Moderate agreement 0.61 – 0.80 Substantial agreement 0.81 – 1.00 Almost perfect agreement November 1, 2011 Crowdsourcing for Research and Engineering 117
Detection Theory • Sensitivity measures – High sensitivity: good ability to discriminate – Low sensitivity: poor ability Stimulus “Yes” “No” Class S1 Hits Misses S2 False alarms Correct rejections Hit rate H = P(“yes”|S2) False alarm rate F = P(“yes”|S1) November 1, 2011 Crowdsourcing for Research and Engineering 118
Finding Consensus • When multiple workers disagree on the correct label, how do we resolve this? – Simple majority vote (or average and round) – Weighted majority vote (e.g. naive bayes) • Many papers from machine learning… • If wide disagreement, likely there is a bigger problem which consensus doesn’t address November 1, 2011 Crowdsourcing for Research and Engineering 120
Quality Control on MTurk • Rejecting work & Blocking workers (more later…) – Requestors don’t want bad PR or complaint emails – Common practice: always pay, block as needed • Approval rate: easy to use, but value? – P. Ipeirotis. Be a Top Mechanical Turk Worker: You Need $5 and 5 Minutes. Oct. 2010 – Many requestors don’t ever reject… • Qualification test – Pre-screen workers’ capabilities & effectiveness – Example and pros/cons in next slides… • Geographic restrictions • Mechanical Turk Masters (June 23, 2011) – Recent addition, degree of benefit TBD… November 1, 2011 Crowdsourcing for Research and Engineering 121
Tools and Packages for MTurk • QA infrastructure layers atop MTurk promote useful separation-of-concerns from task – TurkIt • Quik Turkit provides nearly realtime services – Turkit-online (??) – Get Another Label (& qmturk) – Turk Surveyor – cv-web-annotation-toolkit (image labeling) – Soylent – Boto (python library) • Turkpipe: submit batches of jobs using the command line. • More needed… November 1, 2011 Crowdsourcing for Research and Engineering 122
A qualification test snippet <Question> <QuestionIdentifier>question1</QuestionIdentifier> <QuestionContent> <Text>Carbon monoxide poisoning is</Text> </QuestionContent> <AnswerSpecification> <SelectionAnswer> <StyleSuggestion>radiobutton</StyleSuggestion> <Selections> <Selection> <SelectionIdentifier>1</SelectionIdentifier> <Text>A chemical technique</Text> </Selection> <Selection> <SelectionIdentifier>2</SelectionIdentifier> <Text>A green energy treatment</Text> </Selection> <Selection> <SelectionIdentifier>3</SelectionIdentifier> <Text>A phenomena associated with sports</Text> </Selection> <Selection> <SelectionIdentifier>4</SelectionIdentifier> <Text>None of the above</Text> </Selection> </Selections> </SelectionAnswer> </AnswerSpecification> </Question> 2011 November 1, Crowdsourcing for Research and Engineering 123
Qualification tests: pros and cons • Advantages – Great tool for controlling quality – Adjust passing grade • Disadvantages – Extra cost to design and implement the test – May turn off workers, hurt completion time – Refresh the test on a regular basis – Hard to verify subjective tasks like judging relevance • Try creating task-related questions to get worker familiar with task before starting task in earnest November 1, 2011 Crowdsourcing for Research and Engineering 124
More on quality control & assurance • HR issues: recruiting, selection, & retention – e.g., post/tweet, design a better qualification test, bonuses, … • Collect more redundant judgments… – at some point defeats cost savings of crowdsourcing – 5 workers is often sufficient November 1, 2011 Crowdsourcing for Research and Engineering 125
Robots and Captchas • Some reports of robots on MTurk – E.g. McCreadie et al. (2011) – violation of terms of service – Artificial artificial artificial intelligence • Captchas seem ideal, but… – There is abuse of robots using turkers to solve captchas so they can access web resources – Turker wisdom is therefore to avoid such HITs • What to do? – Use standard captchas, notify workers – Block robots other ways (e.g. external HITs) – Catch robots through standard QC, response times – Use HIT-specific captchas (Kazai et al., 2011) November 1, 2011 Crowdsourcing for Research and Engineering 126
Was the task difficult? • Ask workers to rate difficulty of a search topic • 50 topics; 5 workers, $0.01 per task November 1, 2011 Crowdsourcing for Research and Engineering 127
Other quality heuristics • Justification/feedback as quasi-captcha – Successfully proven in past experiments – Should be optional – Automatically verifying feedback was written by a person may be difficult (classic spam detection task) • Broken URL/incorrect object – Leave an outlier in the data set – Workers will tell you – If somebody answers “excellent” on a graded relevance test for a broken URL => probably spammer November 1, 2011 Crowdsourcing for Research and Engineering 128
Dealing with bad workers • Pay for “bad” work instead of rejecting it? – Pro: preserve reputation, admit if poor design at fault – Con: promote fraud, undermine approval rating system • Use bonus as incentive – Pay the minimum $0.01 and $0.01 for bonus – Better than rejecting a $0.02 task • If spammer “caught”, block from future tasks – May be easier to always pay, then block as needed November 1, 2011 Crowdsourcing for Research and Engineering 129
Worker feedback • Real feedback received via email after rejection • Worker XXX I did. If you read these articles most of them have nothing to do with space programs. I’m not an idiot. • Worker XXX As far as I remember there wasn't an explanation about what to do when there is no name in the text. I believe I did write a few comments on that, too. So I think you're being unfair rejecting my HITs. November 1, 2011 Crowdsourcing for Research and Engineering 130
Real email exchange with worker after rejection WORKER: this is not fair , you made me work for 10 cents and i lost my 30 minutes of time ,power and lot more and gave me 2 rejections at least you may keep it pending. please show some respect to turkers REQUESTER: I'm sorry about the rejection. However, in the directions given in the hit, we have the following instructions: IN ORDER TO GET PAID, you must judge all 5 webpages below *AND* complete a minimum of three HITs. Unfortunately, because you only completed two hits, we had to reject those hits. We do this because we need a certain amount of data on which to make decisions about judgment quality. I'm sorry if this caused any distress. Feel free to contact me if you have any additional questions or concerns. WORKER: I understood the problems. At that time my kid was crying and i went to look after. that's why i responded like that. I was very much worried about a hit being rejected. The real fact is that i haven't seen that instructions of 5 web page and started doing as i do the dolores labs hit, then someone called me and i went to attend that call. sorry for that and thanks for your kind concern. November 1, 2011 Crowdsourcing for Research and Engineering 131
Exchange with worker • Worker XXX Thank you. I will post positive feedback for you at Turker Nation. Me: was this a sarcastic comment? • I took a chance by accepting some of your HITs to see if you were a trustworthy author. My experience with you has been favorable so I will put in a good word for you on that website. This will help you get higher quality applicants in the future, which will provide higher quality work, which might be worth more to you, which hopefully means higher HIT amounts in the future. November 1, 2011 Crowdsourcing for Research and Engineering 132
Build Your Reputation as a Requestor • Word of mouth effect – Workers trust the requester (pay on time, clear explanation if there is a rejection) – Experiments tend to go faster – Announce forthcoming tasks (e.g. tweet) • Disclose your real identity? November 1, 2011 Crowdsourcing for Research and Engineering 133
Other practical tips • Sign up as worker and do some HITs • “Eat your own dog food” • Monitor discussion forums • Address feedback (e.g., poor guidelines, payments, passing grade, etc.) • Everything counts! – Overall design only as strong as weakest link November 1, 2011 Crowdsourcing for Research and Engineering 134
Content quality • People like to work on things that they like • TREC ad-hoc vs. INEX – TREC experiments took twice to complete – INEX (Wikipedia), TREC (LA Times, FBIS) • Topics – INEX: Olympic games, movies, salad recipes, etc. – TREC: cosmic events, Schengen agreement, etc. • Content and judgments according to modern times – Airport security docs are pre 9/11 – Antarctic exploration (global warming ) November 1, 2011 Crowdsourcing for Research and Engineering 135
Content quality - II • Document length • Randomize content • Avoid worker fatigue – Judging 100 documents on the same subject can be tiring, leading to decreasing quality November 1, 2011 Crowdsourcing for Research and Engineering 136
Presentation • People scan documents for relevance cues • Document design • Highlighting no more than 10% November 1, 2011 Crowdsourcing for Research and Engineering 137
Presentation - II November 1, 2011 Crowdsourcing for Research and Engineering 138
Relevance justification • Why settle for a label? • Let workers justify answers – cf. Zaidan et al. (2007) “annotator rationales” • INEX – 22% of assignments with comments • Must be optional • Let’s see how people justify November 1, 2011 Crowdsourcing for Research and Engineering 139
“Relevant” answers [Salad Recipes] Doesn't mention the word 'salad', but the recipe is one that could be considered a salad, or a salad topping, or a sandwich spread. Egg salad recipe Egg salad recipe is discussed. History of salad cream is discussed. Includes salad recipe It has information about salad recipes. Potato Salad Potato salad recipes are listed. Recipe for a salad dressing. Salad Recipes are discussed. Salad cream is discussed. Salad info and recipe The article contains a salad recipe. The article discusses methods of making potato salad. The recipe is for a dressing for a salad, so the information is somewhat narrow for the topic but is still potentially relevant for a researcher. This article describes a specific salad. Although it does not list a specific recipe, it does contain information relevant to the search topic. gives a recipe for tuna salad relevant for tuna salad recipes relevant to salad recipes this is on-topic for salad recipes November 1, 2011 Crowdsourcing for Research and Engineering 140
“Not relevant” answers [Salad Recipes] About gaming not salad recipes. Article is about Norway. Article is about Region Codes. Article is about forests. Article is about geography. Document is about forest and trees. Has nothing to do with salad or recipes. Not a salad recipe Not about recipes Not about salad recipes There is no recipe, just a comment on how salads fit into meal formats. There is nothing mentioned about salads. While dressings should be mentioned with salads, this is an article on one specific type of dressing, no recipe for salads. article about a swiss tv show completely off-topic for salad recipes not a salad recipe not about salad recipes totally off base November 1, 2011 Crowdsourcing for Research and Engineering 141
Feedback length • Workers will justify answers • Has to be optional for good feedback • In E51, mandatory comments – Length dropped – “Relevant” or “Not Relevant November 1, 2011 Crowdsourcing for Research and Engineering 143
Other design principles • Text alignment • Legibility • Reading level: complexity of words and sentences • Attractiveness (worker’s attention & enjoyment) • Multi-cultural / multi-lingual • Who is the audience (e.g. target worker community) – Special needs communities (e.g. simple color blindness) • Parsimony • Cognitive load: mental rigor needed to perform task • Exposure effect November 1, 2011 Crowdsourcing for Research and Engineering 144
Platform alternatives • Why MTurk – Amazon brand, lots of research papers – Speed, price, diversity, payments • Why not – Crowdsourcing != Mturk – Spam, no analytics, must build tools for worker & task quality • How to build your own crowdsourcing platform – Back-end – Template language for creating experiments – Scheduler – Payments? November 1, 2011 Crowdsourcing for Research and Engineering 145
The human side • As a worker – I hate when instructions are not clear – I’m not a spammer – I just don’t get what you want – Boring task – A good pay is ideal but not the only condition for engagement • As a requester – Attrition – Balancing act: a task that would produce the right results and is appealing to workers – I want your honest answer for the task – I want qualified workers; system should do some of that for me • Managing crowds and tasks is a daily activity – more difficult than managing computers November 1, 2011 Crowdsourcing for Research and Engineering 146
Things that work • Qualification tests • Honey-pots • Good content and good presentation • Economy of attention • Things to improve – Manage workers in different levels of expertise including spammers and potential cases. – Mix different pools of workers based on different profile and expertise levels. November 1, 2011 Crowdsourcing for Research and Engineering 147
Things that need work • UX and guidelines – Help the worker – Cost of interaction • Scheduling and refresh rate • Exposure effect • Sometimes we just don’t agree • How crowdsourcable is your task November 1, 2011 Crowdsourcing for Research and Engineering 148
V. FUTURE TRENDS: FROM LABELING TO HUMAN COMPUTATION November 1, 2011 Crowdsourcing for Research and Engineering 149
The Turing Test (Alan Turing, 1950) November 1, 2011 Crowdsourcing for Research and Engineering 150
The Turing Test (Alan Turing, 1950) November 1, 2011 Crowdsourcing for Research and Engineering 152
What is a Computer? November 1, 2011 Crowdsourcing for Research and Engineering 153
• What was old is new • “Crowdsourcing: A New Branch of Computer Science” (March 29, 2011) • See also: M. Croarken (2003), Tabulating the heavens: computing the Nautical Almanac in 18th-century England Princeton University Press, 2005 November 1, 2011 Crowdsourcing for Research and Engineering 154
Davis et al. (2010) The HPU. HPU November 1, 2011 Crowdsourcing for Research and Engineering 155
Remembering the Human in HPU • Not just turning a mechanical crank November 1, 2011 Crowdsourcing for Research and Engineering 156
Human Computation Rebirth of people as ‘computists’; people do tasks computers cannot (do well) Stage 1: Detecting robots – CAPTCHA: Completely Automated Public Turing test to tell Computers and Humans Apart – No useful work produced; people just answer questions with known answers Stage 2: Labeling data (at scale) – E.g. ESP game, typical use of MTurk – Game changer for AI: starving for data Stage 3: General “human computation” (HPU) – people do arbitrarily sophisticated tasks (i.e. compute arbitrary functions) – HPU as core component in system architecture, many “HPC” invocations – blend HPU with automation for a new class of hybrid applications – New tradeoffs possible in latency/cost vs. functionality/accuracy November 1, 2011 Crowdsourcing for Research and Engineering 157
Mobile Phone App: “Amazon Remembers” November 1, 2011 Crowdsourcing for Research and Engineering 158
Soylent: A Word Processor with a Crowd Inside • Bernstein et al., UIST 2010 November 1, 2011 Crowdsourcing for Research and Engineering 159
CrowdSearch and mCrowd • T. Yan, MobiSys 2010 November 1, 2011 Crowdsourcing for Research and Engineering 160
Translation by monolingual speakers • C. Hu, CHI 2009 November 1, 2011 Crowdsourcing for Research and Engineering 161
Wisdom of Crowds (WoC) Requires • Diversity • Independence • Decentralization • Aggregation Input: large, diverse sample (to increase likelihood of overall pool quality) Output: consensus or selection (aggregation) November 1, 2011 Crowdsourcing for Research and Engineering 162
WoC vs. Ensemble Learning • Combine multiple models to improve performance over any constituent model – Can use many weak learners to make a strong one – Compensate for poor models with extra computation • Works better with diverse, independent learners • cf. NIPS 2010-2011 Workshops – Computational Social Science & the Wisdom of Crowds • More investigation needed of traditional feature- based machine learning & ensemble methods for consensus labeling with crowdsourcing November 1, 2011 Crowdsourcing for Research and Engineering 163
Unreasonable Effectiveness of Data • Massive free Web data changed how we train learning systems – Banko and Brill (2001). Human Language Tech. – Halevy et al. (2009). IEEE Intelligent Systems. • How might access to cheap & plentiful labeled data change the balance again? November 1, 2011 Crowdsourcing for Research and Engineering 164
Active Learning • Minimize number of labels to achieve goal accuracy rate of classifier – Select examples to label to maximize learning • Vijayanarasimhan and Grauman (CVPR 2011) – Simple margin criteria: select maximally uncertain examples to label next – Finding which examples are uncertain can be computationally intensive (workers have to wait) – Use locality-sensitive hashing to find uncertain examples in sub-linear time November 1, 2011 Crowdsourcing for Research and Engineering 165
Active Learning (2) • V&G report each learning iteration ~ 75 min – 15 minutes for model training & selection – 60 minutes waiting for crowd labels • Leaving workers idle may lose them, slowing uptake and completion times • Keep workers occupied – Mason and Suri (2010): paid waiting room – Laws et al. (EMNLP 2011): parallelize labeling and example selection via producer-consumer model • Workers consume examples, produce labels • Model consumes label, produces examples November 1, 2011 Crowdsourcing for Research and Engineering 166
MapReduce with human computation • Commonalities – Large task divided into smaller sub-problems – Work distributed among worker nodes (workers) – Collect all answers and combine them – Varying performance of heterogeneous CPUs/HPUs • Variations – Human response latency / size of “cluster” – Some tasks are not suitable November 1, 2011 Crowdsourcing for Research and Engineering 167
A Few Questions • How should we balance automation vs. human computation? Which does what? • Who’s the right person for the job? • How do we handle complex tasks? Can we decompose them into smaller tasks? How? November 1, 2011 Crowdsourcing for Research and Engineering 168
Research problems – operational • Methodology – Budget, people, document, queries, presentation, incentives, etc. – Scheduling – Quality • What’s the best “mix” of HC for a task? • What are the tasks suitable for HC? • Can I crowdsource my task? – Eickhoff and de Vries, WSDM 2011 CSDM Workshop November 1, 2011 Crowdsourcing for Research and Engineering 169
More problems • Human factors vs. outcomes • Editors vs. workers • Pricing tasks • Predicting worker quality from observable properties (e.g. task completion time) • HIT / Requestor ranking or recommendation • Expert search : who are the right workers given task nature and constraints • Ensemble methods for Crowd Wisdom consensus November 1, 2011 Crowdsourcing for Research and Engineering 170
Problems: crowds, clouds and algorithms • Infrastructure – Current platforms are very rudimentary – No tools for data analysis • Dealing with uncertainty (propagate rather than mask) – Temporal and labeling uncertainty – Learning algorithms – Search evaluation – Active learning (which example is likely to be labeled correctly) • Combining CPU + HPU – Human Remote Call? – Procedural vs. declarative? – Integration points with enterprise systems November 1, 2011 Crowdsourcing for Research and Engineering 171
CrowdForge: MapReduce for Automation + Human Computation Kittur et al., CHI 2011 November 1, 2011 Crowdsourcing for Research and Engineering 172
Conclusions • Crowdsourcing works and is here to stay • Fast turnaround, easy to experiment, cheap • Still have to design the experiments carefully! • Usability considerations • Worker quality • User feedback extremely useful November 1, 2011 Crowdsourcing for Research and Engineering 173
Conclusions - II • Lots of opportunities to improve current platforms • Integration with current systems • While MTurk first to-market in micro-task vertical, many other vendors are emerging with different affordances or value-added features • Many open research problems … November 1, 2011 Crowdsourcing for Research and Engineering 174
Conclusions – III • Important to know your limitations and be ready to collaborate • Lots of different skills and expertise required – Social/behavioral science – Human factors – Algorithms – Economics – Distributed systems – Statistics November 1, 2011 Crowdsourcing for Research and Engineering 175
VIII REFERENCES & RESOURCES November 1, 2011 Crowdsourcing for Research and Engineering 176
Books • Omar Alonso, Gabriella Kazai, and Stefano Mizzaro. (2012). Crowdsourcing for Search Engine Evaluation: Why and How. • Law and von Ahn (2011). Human Computation November 1, 2011 Crowdsourcing for Research and Engineering 177
More Books July 2010, kindle-only: “This book introduces you to the top crowdsourcing sites and outlines step by step with photos the exact process to get started as a requester on Amazon Mechanical Turk.“ November 1, 2011 Crowdsourcing for Research and Engineering 178
2011 Tutorials a

Crowdsourcing For Research and Engineering (Tutorial given at CrowdConf 2011)

Check these out next

Crowdsourcing For Research and Engineering (Tutorial given at CrowdConf 2011)

Recommended

More Related Content

Slideshows for you(7)

Viewers also liked(20)

Similar to Crowdsourcing For Research and Engineering (Tutorial given at CrowdConf 2011)(20)

More from Matthew Lease(18)

Recently uploaded(20)

Crowdsourcing For Research and Engineering (Tutorial given at CrowdConf 2011)