{"id":"W4206369543","doi":"10.2139/ssrn.3989656","title":"Learning True Labels from Noisy Crowd-Sourced Annotations","year":2021,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Mobile Crowdsensing and Crowdsourcing","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Computer science; Crowdsourcing; Crowd sourcing; Artificial intelligence; Information retrieval; Natural language processing; Data science; World Wide Web","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0009993946,0.0001947569,0.000233821,0.0001102591,0.0007773705,0.0005974877,0.0005504495,0.00009453671,0.0000551013],"category_scores_gemma":[0.00025373,0.0001969441,0.0001638881,0.0005568378,0.00003592482,0.0004258047,0.0001361309,0.002388591,0.000133931],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000415131,"about_ca_system_score_gemma":0.002186292,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007864813,"about_ca_topic_score_gemma":0.0002735399,"domain_scores_codex":[0.9966483,0.0002957834,0.0003511927,0.000426181,0.0003708418,0.001907695],"domain_scores_gemma":[0.9988024,0.0001900177,0.0001913239,0.0004229512,0.0002437145,0.0001495943],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00003435372,0.0003025441,0.004120307,0.00001407175,0.0007476131,0.0004384248,0.009012874,0.01396108,0.08412167,0.3137212,0.001314712,0.5722111],"study_design_scores_gemma":[0.00678245,0.001318774,0.007595848,0.0004628327,0.0002916292,0.01035893,0.02027168,0.1143242,0.04323488,0.6700783,0.1222548,0.003025645],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4798883,0.002496788,0.5137568,0.001851687,0.0003877298,0.00004340477,0.000001209992,0.0001716777,0.001402315],"genre_scores_gemma":[0.9887624,0.0004314577,0.004771739,0.0002908705,0.0004063144,0.000002938828,0.000008431861,0.00002657015,0.0052993],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5691855,"threshold_uncertainty_score":0.9999129,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.005698329588986957,"score_gpt":0.2147317712613018,"score_spread":0.2090334416723148,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}