{"id":"W2972483465","doi":"10.2196/14830","title":"Fine-Tuning Bidirectional Encoder Representations From Transformers (BERT)–Based Models on Large-Scale Electronic Health Record Notes: An Empirical Study","year":2019,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":175,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Heart, Lung, and Blood Institute; U.S. Department of Veterans Affairs","keywords":"Normalization (sociology); Computer science; Natural language processing; Named-entity recognition; Artificial intelligence; SNOMED CT; Transformer; Encoder; Language model; Information retrieval; Health records; Electronic health record; Machine learning; Task (project management); Health care; Linguistics; Terminology","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":[0.001271333,0.0002623808,0.0004193122,0.0002623851,0.0003168323,0.0001205712,0.0008935378,0.0002031376,0.0003804799],"category_scores_gemma":[0.0001402801,0.000230055,0.0001114307,0.0006606716,0.00004212408,0.0009654575,0.0001067666,0.001394664,0.0001491079],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002892932,"about_ca_system_score_gemma":0.001496106,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000662384,"about_ca_topic_score_gemma":0.001577355,"domain_scores_codex":[0.9956073,0.0004211962,0.001047682,0.0004072821,0.001725767,0.0007907955],"domain_scores_gemma":[0.9973677,0.0008073308,0.0002791842,0.0007923749,0.0001204452,0.0006329571],"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":[0.0003461954,0.005765596,0.3067727,0.0005980383,0.0002305933,0.00002501595,0.2637541,0.09083769,0.000006233219,0.004687914,0.00646803,0.3205079],"study_design_scores_gemma":[0.001359873,0.001363571,0.006813373,0.00009600774,0.000004757541,0.000004629826,0.001941036,0.9849116,0.000004230589,0.0006881669,0.002575147,0.0002376679],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4348921,0.00001644606,0.556284,0.006170683,0.0003909167,0.0009440218,0.00001838349,0.00034014,0.0009433262],"genre_scores_gemma":[0.964292,0.00001124929,0.02689104,0.008243214,0.0001436545,0.0001533282,0.0001630395,0.00002740927,0.00007510203],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8940738,"threshold_uncertainty_score":0.9381369,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03612909331573435,"score_gpt":0.3803868410582724,"score_spread":0.3442577477425381,"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."}}