{"id":"W2090601220","doi":"10.1145/2660168.2660186","title":"Correcting Large-Scale OMR Data with Crowdsourcing","year":2014,"lang":"en","type":"article","venue":"","topic":"Mobile Crowdsensing and Crowdsourcing","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Crowdsourcing; Computer science; Variety (cybernetics); Scale (ratio); Data science; Process (computing); Globe; Ground truth; Reliability (semiconductor); World Wide Web; Human–computer interaction; Artificial intelligence","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01001936,0.001643423,0.002067474,0.004203775,0.002506445,0.003299305,0.003405164,0.002183805,0.003256286],"category_scores_gemma":[0.03681767,0.0008281561,0.001526211,0.003884947,0.002406092,0.002969019,0.007455542,0.002111308,0.002865451],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001630515,"about_ca_system_score_gemma":0.002528619,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01169273,"about_ca_topic_score_gemma":0.01514949,"domain_scores_codex":[0.987699,0.003786372,0.0005311394,0.002653037,0.004754064,0.0005762583],"domain_scores_gemma":[0.9709011,0.01048423,0.002238411,0.01148223,0.00434366,0.0005504034],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0008440571,0.0003693394,0.01966954,0.001364677,0.0007462934,0.00166065,0.003954835,0.281642,0.05349628,0.01802624,0.03266477,0.5855614],"study_design_scores_gemma":[0.0001814432,0.0002920844,0.01481263,0.0003632891,0.0001484408,0.0006835565,0.004065796,0.7747408,0.03253976,0.08345298,0.08836506,0.0003542075],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04739524,0.0007190445,0.9304064,0.002245193,0.0008389867,0.0006962001,0.00171559,0.00664718,0.009336189],"genre_scores_gemma":[0.4228864,0.0004072862,0.5640211,0.0009699516,0.0004699121,0.0005105651,0.00290479,0.001040748,0.006789307],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01169273,"threshold_uncertainty_score":0.05298811,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01433183667179794,"score_gpt":0.2316893982502087,"score_spread":0.2173575615784107,"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."}}