{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002544954,0.00009240752,0.0001099461,0.00003688918,0.00008116816,0.0003156692,0.0002340476,0.00004752051,0.000002240478],"category_scores_gemma":[0.00007688158,0.00007673082,0.00002670214,0.0001866311,0.000004814101,0.0002196698,0.0001199045,0.00004764568,0.000009498752],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003556227,"about_ca_system_score_gemma":0.00004798969,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000271137,"about_ca_topic_score_gemma":0.000005576849,"domain_scores_codex":[0.9991479,0.0000530472,0.000142521,0.0004021902,0.0001025143,0.0001518343],"domain_scores_gemma":[0.9993384,0.00003600071,0.0000258241,0.0002226062,0.00006034721,0.0003167815],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00006941133,0.00002478997,0.001307031,0.001455187,0.00004398813,0.00007766309,0.02132447,0.01751121,0.005926008,0.9006151,0.03056467,0.02108051],"study_design_scores_gemma":[0.000296142,0.00008482202,0.00002893573,0.00003838123,0.000004572072,0.00001560886,0.00006558306,0.965654,0.0008207277,0.0003409226,0.03249732,0.0001530287],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01213041,0.000106057,0.9660335,0.02070133,0.0002131785,0.0003236085,0.000006097208,0.0003087411,0.0001770814],"genre_scores_gemma":[0.65258,0.000003463413,0.3306353,0.01614735,0.0002121277,0.00005015671,0.000005405225,0.000008061042,0.0003581612],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9481428,"threshold_uncertainty_score":0.3128992,"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."}}