{"id":"W2976978765","doi":"10.2196/13430","title":"Impact of Automatic Query Generation and Quality Recognition Using Deep Learning to Curate Evidence From Biomedical Literature: Empirical Study","year":2019,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Quality (philosophy); Artificial intelligence; Natural language processing; Deep learning; Information retrieval; Data science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008699453,0.0001756352,0.0003351498,0.00009328449,0.00005579584,0.00005191779,0.0001751319,0.0003874383,0.00009372227],"category_scores_gemma":[0.001642153,0.0001284309,0.00008581041,0.0002500713,0.0001424255,0.00002593237,0.0001954492,0.0003050184,0.00001873822],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003236755,"about_ca_system_score_gemma":0.0001625253,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007232114,"about_ca_topic_score_gemma":0.00001534981,"domain_scores_codex":[0.9979625,0.000261023,0.0007547133,0.0002032463,0.0005876666,0.0002308526],"domain_scores_gemma":[0.99892,0.0001632232,0.000233731,0.0002426799,0.0001257287,0.0003146317],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0003426055,0.0008857998,0.394889,0.0005942607,0.0004421466,0.00001367829,0.03348852,0.00009580611,0.07824184,0.000001093753,0.001145129,0.4898601],"study_design_scores_gemma":[0.00502644,0.01238317,0.5540867,0.002776924,0.0001704317,0.0001032593,0.0132757,0.4053159,0.004433941,0.00006373726,0.0009952622,0.001368461],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.993643,0.000299265,0.005433673,0.0000714347,0.0001359222,0.0003569173,0.00001786679,0.00002952122,0.00001241406],"genre_scores_gemma":[0.9917157,0.00007766423,0.007544652,0.0001972946,0.0001840044,0.00001796875,0.0002434202,0.00000985424,0.000009465963],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4884917,"threshold_uncertainty_score":0.523726,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08052265100987234,"score_gpt":0.4162881156968011,"score_spread":0.3357654646869288,"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."}}