{"id":"W4381377622","doi":"10.1101/2023.06.18.23291567","title":"Machine learning to increase the efficiency of a literature surveillance system: a performance evaluation","year":2023,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Computer science; Machine learning; Critical appraisal; Artificial intelligence; MEDLINE; Set (abstract data type); Sensitivity (control systems); Binary classification; Information retrieval; Medical physics; Medicine; Support vector machine; Alternative medicine; Pathology","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":[0.002197617,0.0001801787,0.0002354615,0.0000683714,0.00008817683,0.00002820715,0.000453628,0.0002872552,0.000003849335],"category_scores_gemma":[0.001454377,0.0001174744,0.00009514858,0.0002335239,0.00008801581,7.458672e-7,0.0005725768,0.0003719028,0.00001237762],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001957313,"about_ca_system_score_gemma":0.0001212836,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008151771,"about_ca_topic_score_gemma":0.00003712844,"domain_scores_codex":[0.9983894,0.0003735532,0.000275332,0.000417248,0.0003485263,0.0001959783],"domain_scores_gemma":[0.998952,0.00005793707,0.0001774848,0.0005311151,0.0002215039,0.00005992828],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001023723,0.0002827579,0.6590303,0.006030383,0.0006727691,0.000043031,0.005558801,0.02248893,0.175847,0.00007111115,0.004832863,0.1241184],"study_design_scores_gemma":[0.002572123,0.002778667,0.6601889,0.006260527,0.0003485463,0.0001171548,0.001301023,0.2119904,0.04139194,0.00009200007,0.07077182,0.002186946],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9935651,0.004535816,0.0004182108,0.0003095611,0.0004493237,0.0003438323,0.00005083411,0.00005105669,0.000276255],"genre_scores_gemma":[0.9982641,0.0003409697,0.000223128,0.00003615353,0.0001925668,0.0001089804,0.0002812133,0.00002018863,0.0005326507],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1895014,"threshold_uncertainty_score":0.4790466,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02354627931456999,"score_gpt":0.2870682422268218,"score_spread":0.2635219629122518,"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."}}