quality control part 2
play

Quality Control - part 2 Crowdsourcing and Human Computation - PowerPoint PPT Presentation

Quality Control - part 2 Crowdsourcing and Human Computation Instructor: Chris Callison-Burch Website: crowdsourcing-class.org Different Mechanisms for Quality Control Aggregation and redundancy Embedded gold standard data


  1. Expectation Maximization algorithm • EM is an algorithm for finding the probabilities of unobserved variables • We will use it to estimate how accurate workers’ labels are, and infer how good each worker is • This is more sophisticated than voting

  2. Dawid and Skene (1977) • Maximum Likelihood Estimation of Observer Error-rates using the EM Algorithm • Examined application to medical diagnosis • Patients are sometimes treated by multiple physicians, who can give different diagnoses • Why? Doctors may has different questions. Patient may describe history differently. Doctors may classify symptoms differently

  3. Observer Error • Given that different doctors have different opinions, they can’t all be right. • How often do individual physicians suffer from “observer error”? Are their errors systematic? • Answers depend on the “true” diagnosis

  4. Observer Error • Observer error would be easy to calculate if we had ground truth • Simply count the misdiagnoses and divide by the total number of diagnoses • However, sometimes it is impossible to know what diagnosis is correct. Same set of symptoms can arise from multiple root causes.

  5. “I know it when I see it” I shall not today attempt further to define "hard- core pornography"; and perhaps I could never succeed in intelligibly doing so. But I know it when I see it. —Justice Potter Stewart

  6. url worker1 worker2 worker3 worker4 worker5 porn not not not porn sunnyfun.com porn porn porn porn porn sex-mission.com not porn not not porn google.com porn porn porn porn not youporn.com porn not not not porn yahoo.com

  7. Solution? • Can’t have Justice Stewart rule on everything • Instead, we will apply Dawid and Skene’s EM algorithm, which iteratively 1. Estimates the correct answers, using labels from multiple workers, and accounts for the quality of each worker 2. Estimates the quality of the workers by comparing the submitted answers to the inferred correct answers

  8. Inputs • a set of N objects o 1 ... o N 
 sunnyfun.com, sex-mission.com, google.com, youporn.com, yahoo.com • a set of L possible labels: 
 {porn, not porn} • Labels for each object by K workers 
 worker1, worker2, worker3, worker4, worker5

  9. Goal 1 • Recover the true class label T (o n ) for each object o n when “gold” truth is unknown • Since the true labels are not known / never directly observed, they are called latent variables

  10. Goal 2 • For each worker who contributed labels, calculate their accuracy or reliability • To calculate accuracy show how often they mistakenly choose one label when a different one is the actual truth

  11. Chicken and egg problem • If we knew what the true class labels were for each object for each object, then we could compute each Turker’s accuracy • If we had accuracies for every Turker , then we could infer what the true label for each object should be

  12. Input : Labels l [ k ][ n ] from worker ( k ) to object o n , Output : Confusion matrix π ( k ) for each worker ( k ), Correct ij labels T ( o n ) for each object o n , Class priors Pr { C } for each class C 1 Initialize error rates π ( k ) for each worker ( k ) (e.g., assume each ij worker is perfect); 2 Initialize correct label for each object T ( o n ) (e.g., using majority vote); 3 while not converged do Estimate the correct label T ( o n ) for each object, using the 4 labels l [ · ][ n ] assigned to o n by workers, weighting the votes using the error rates π ( k ) ij ; Estimate the error rates π ( k ) ij , for each worker ( k ), using 5 the correct labels T ( o n ) and the assigned labels l [ k ][ n ]; Estimate the class priors Pr { C } , for each class C ; 6 7 end 8 return Estimated error rates π ( k ) ij , Estimated correct labels T ( o n ) , Estimated class priors Pr { C } Algorithm 1: The EM algorithm for worker quality estima- tion.

  13. Input url worker1 worker2 worker3 worker4 worker5 porn not not not porn sunnyfun.com porn porn porn porn porn sex-mission.com not porn not not porn google.com porn porn porn porn not youporn.com porn not not not porn yahoo.com

  14. Output: “True” Labels True url Labels not sunnyfun.com sex- porn mission.com not google.com porn youporn.com not yahoo.com

  15. Output: Worker Accuracies Guess not w1 porn porn 100% 0% Truth porn Guess not 67% 33% not w4 porn porn porn 100% 0% Truth porn not w2 porn not 0% 100% porn 100% 0% porn porn not 33% 67% not w5 porn porn porn 50% 50% porn not w3 porn not 100% 0% porn 100% 0% porn porn not 0% 100% porn

  16. Initialize Guess not w1 porn porn 100% 0% Truth porn Guess not 0% 100% not w4 porn porn porn 100% 0% Truth porn not w2 porn not 0% 100% porn 100% 0% porn porn not 0% 100% not w5 porn porn porn 100% 0% porn not w3 porn not 0% 100% porn 100% 0% porn porn not 0% 100% porn

  17. Initialize with majority vote worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 40% 60% sex- porn porn porn porn porn 100% 0% mission.com not porn not not porn google.com 40% 60% youporn.com porn porn porn porn not 80% 20% porn not not not porn yahoo.com 40% 60%

  18. Pretend your labels are correct worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0% 100% sex- porn porn porn porn porn 100% 0% mission.com not porn not not porn google.com 0% 100% youporn.com porn porn porn porn not 100% 0% porn not not not porn yahoo.com 0% 100%

  19. Guess Re-Calculate not w1 porn Worker porn 0 0 Truth porn Scores not 1 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  20. Guess Re-Calculate not w1 porn Worker porn 1 0 Truth porn Scores not 1 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  21. Guess Re-Calculate not w1 porn Worker porn 1 0 Truth porn Scores not 1 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  22. Guess Re-Calculate not w1 porn Worker porn 2 0 Truth porn Scores not 1 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  23. Guess Re-Calculate not w1 porn Worker porn 2 0 Truth porn Scores not 2 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  24. Guess Re-Calculate not w1 porn Worker porn 100% 0% Truth porn Scores not 67% 33% porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  25. Guess Re-Calculate not w2 porn Worker porn 0 0 Truth porn Scores not 0 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  26. Guess Re-Calculate not w2 porn Worker porn 1 0 Truth porn Scores not 0 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  27. Guess Re-Calculate not w2 porn Worker porn 1 0 Truth porn Scores not 1 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  28. Guess Re-Calculate not w2 porn Worker porn 2 0 Truth porn Scores not 1 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  29. Guess Re-Calculate not w2 porn Worker porn 2 0 Truth porn Scores not 1 2 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  30. Guess Re-Calculate not w2 porn Worker porn 100% 0% Truth porn Scores not 33% 67% porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  31. Guess Re-Calculate not w3 porn Worker porn 0 0 Truth porn Scores not 0 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  32. Guess Re-Calculate not w3 porn Worker porn 1 0 Truth porn Scores not 0 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  33. Guess Re-Calculate not w3 porn Worker porn 1 0 Truth porn Scores not 0 2 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  34. Guess Re-Calculate not w3 porn Worker porn 2 0 Truth porn Scores not 0 2 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  35. Guess Re-Calculate not w3 porn Worker porn 2 0 Truth porn Scores not 0 3 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  36. Guess Re-Calculate not w3 porn Worker porn 100% 0% Truth porn Scores not 0% 100% porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  37. Guess Re-Calculate not w4 porn Worker porn 0 0 Truth porn Scores not 0 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  38. Guess Re-Calculate not w4 porn Worker porn 1 0 Truth porn Scores not 0 1 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  39. Guess Re-Calculate not w4 porn Worker porn 1 0 Truth porn Scores not 0 2 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  40. Guess Re-Calculate not w4 porn Worker porn 2 0 Truth porn Scores not 0 2 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  41. Guess Re-Calculate not w4 porn Worker porn 2 0 Truth porn Scores not 0 3 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  42. Guess Re-Calculate not w4 porn Worker porn 100% 0% Truth porn Scores not 0% 100% porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  43. Guess Re-Calculate not w5 porn Worker porn 0 0 Truth porn Scores not 1 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  44. Guess Re-Calculate not w5 porn Worker porn 1 0 Truth porn Scores not 1 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  45. Guess Re-Calculate not w5 porn Worker porn 1 0 Truth porn Scores not 2 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  46. Guess Re-Calculate not w5 porn Worker porn 1 1 Truth porn Scores not 2 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  47. Guess Re-Calculate not w5 porn Worker porn 1 1 Truth porn Scores not 3 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  48. Guess Re-Calculate not w5 porn Worker porn 1 1 Truth porn Scores not 3 0 porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  49. Guess Re-Calculate not w5 porn Worker porn 50% 50% Truth porn Scores not 100% 0% porn worker worker worker worker worker not url porn 1 2 3 4 5 porn sunnyfun.com porn not not not porn 0 1 sex- porn porn porn porn porn 1 0 mission.com not porn not not porn google.com 0 1 youporn.com porn porn porn porn not 1 0 porn not not not porn yahoo.com 0 1

  50. Guess Update not w1 porn Estimates for porn 1.00 0.00 Truth porn True Labels not 0.67 0.33 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 1.00 0.67 sex- porn porn porn porn porn mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  51. Guess Update not w2 porn Estimates for porn 1.00 0.00 Truth porn True Labels not 0.33 0.67 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 1.00 1.34 sex- porn porn porn porn porn mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  52. Guess Update not w3 porn Estimates for porn 1 0 Truth porn True Labels not 0 1 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 1.00 2.34 sex- porn porn porn porn porn mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  53. Guess Update not w4 porn Estimates for porn 1 0 Truth porn True Labels not 0 1 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 1.00 3.34 sex- porn porn porn porn porn mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  54. Guess Update not w5 porn Estimates for porn 0.5 0.5 Truth porn True Labels not 1 0 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 1.50 4.34 sex- porn porn porn porn porn mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  55. Update Estimates for True Labels not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 1.50 4.34 sex- porn porn porn porn porn mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  56. Update Estimates for True Labels not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  57. Guess Update not w1 porn Estimates for porn 1.00 0.00 Truth porn True Labels not 0.67 0.33 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 1.00 0.67 mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  58. Guess Update not w2 porn Estimates for porn 1.00 0.00 Truth porn True Labels not 0.33 0.67 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 2.00 1.00 mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  59. Guess Update not w3 porn Estimates for porn 1 0 Truth porn True Labels not 0 1 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 3.00 1.00 mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  60. Guess Update not w4 porn Estimates for porn 1 0 Truth porn True Labels not 0 1 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 4.00 1.00 mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  61. Guess Update not w5 porn Estimates for porn 0.5 0.5 Truth porn True Labels not 1 0 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 4.50 2.00 mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  62. Update Estimates for True Labels not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 4.50 2.00 mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  63. Update Estimates for True Labels not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 69% 31% mission.com not porn not not porn google.com youporn.com porn porn porn porn not porn not not not porn yahoo.com

  64. Guess Update not w1 porn Estimates for porn 1.00 0.00 Truth porn True Labels not 0.67 0.33 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 69% 31% mission.com not porn not not porn google.com 0.00 0.33 youporn.com porn porn porn porn not porn not not not porn yahoo.com

  65. Guess Update not w2 porn Estimates for porn 1.00 0.00 Truth porn True Labels not 0.33 0.67 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 69% 31% mission.com not porn not not porn google.com 1.00 0.67 youporn.com porn porn porn porn not porn not not not porn yahoo.com

  66. Guess Update not w3 porn Estimates for porn 1 0 Truth porn True Labels not 0 1 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 69% 31% mission.com not porn not not porn google.com 1.00 1.67 youporn.com porn porn porn porn not porn not not not porn yahoo.com

  67. Guess Update not w4 porn Estimates for porn 1 0 Truth porn True Labels not 0 1 porn not worker worker worker worker worker url porn porn 1 2 3 4 5 sunnyfun.com porn not not not porn 26% 74% sex- porn porn porn porn porn 69% 31% mission.com not porn not not porn google.com 1.00 2.67 youporn.com porn porn porn porn not porn not not not porn yahoo.com

Recommend


More recommend