{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006135539,0.002036104,0.0009786861,0.001442901,0.0006229141,0.001360017,0.002141243,0.001391824,0.002723341],"category_scores_gemma":[0.02103714,0.0007151957,0.001151065,0.001624218,0.0006774082,0.003833461,0.001497114,0.002917629,0.001676586],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003049981,"about_ca_system_score_gemma":0.002539148,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.04063042,"about_ca_topic_score_gemma":0.04292474,"domain_scores_codex":[0.9974777,0.001319501,0.0001732545,0.0006019998,0.0002665148,0.0001611176],"domain_scores_gemma":[0.9824448,0.01392921,0.0004914187,0.001347414,0.001531244,0.0002558592],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001133329,0.001298543,0.01860538,0.0008122419,0.0004886114,0.0003209896,0.0004355569,0.5046804,0.005125763,0.002445493,0.02706958,0.4375841],"study_design_scores_gemma":[0.00008895111,0.0002043761,0.002358831,0.00005288435,0.0001082672,0.00008704392,0.0001229955,0.9890055,0.004182559,0.00155695,0.002193567,0.00003801191],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8049118,0.006765442,0.1536925,0.001830152,0.000609474,0.0006991973,0.008461047,0.01518419,0.007846131],"genre_scores_gemma":[0.8907543,0.001118706,0.08198144,0.0004997967,0.0001179494,0.0003920877,0.02019482,0.0005458899,0.004394924],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04063042,"threshold_uncertainty_score":0.0807879,"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."}}