{"id":"W3103026670","doi":"10.18653/v1/2020.sdp-1.4","title":"A Smart System to Generate and Validate Question Answer Pairs for COVID-19 Literature","year":2020,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; Thomson Reuters (Canada); Queen's University","funders":"","keywords":"Computer science; Annotation; Coronavirus disease 2019 (COVID-19); Domain (mathematical analysis); Transformer; Task (project management); Subject-matter expert; Information retrieval; Data science; Artificial intelligence; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008150267,0.001909504,0.001193735,0.009562546,0.001207542,0.002583347,0.002050726,0.002484971,0.01688812],"category_scores_gemma":[0.02976396,0.0009290405,0.001581882,0.002752431,0.0005614958,0.006101384,0.004482348,0.001766513,0.01755948],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001429728,"about_ca_system_score_gemma":0.002238439,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00209436,"about_ca_topic_score_gemma":0.003589816,"domain_scores_codex":[0.9955423,0.001330466,0.0005564576,0.001679742,0.0007506069,0.0001403304],"domain_scores_gemma":[0.9850748,0.008033361,0.0009652217,0.00271545,0.002499403,0.0007117587],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001410314,0.001080139,0.0201952,0.002851496,0.0003761803,0.001265791,0.003738177,0.007332786,0.05350169,0.01271874,0.2071455,0.6883841],"study_design_scores_gemma":[0.0004200128,0.001072868,0.01485894,0.0004287243,0.0003164548,0.00192114,0.001760393,0.4797606,0.09799321,0.03203943,0.3690946,0.0003336635],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03618982,0.0007826192,0.4979724,0.001510996,0.000547399,0.002835851,0.04279316,0.4071433,0.01022448],"genre_scores_gemma":[0.1484429,0.0003229721,0.7312281,0.0008439401,0.000206465,0.002584046,0.1020046,0.00538965,0.008977222],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01688812,"threshold_uncertainty_score":0.05649638,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03858974180858433,"score_gpt":0.2771664547504922,"score_spread":0.2385767129419079,"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."}}