{"id":"W4385571397","doi":"10.18653/v1/2023.clinicalnlp-1.36","title":"WangLab at MEDIQA-Chat 2023: Clinical Note Generation from Doctor-Patient Conversations using Large Language Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"Health Sciences Centre; University Health Network; Sunnybrook Health Science Centre; Vector Institute; University of Toronto","funders":"Alliance de recherche numérique du Canada","keywords":"Computer science; Task (project management); Scrutiny; Context (archaeology); Language model; Natural language processing; Path (computing); Artificial intelligence; Human–computer interaction; Multimedia; World Wide Web; Programming language","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01438641,0.002752857,0.001737671,0.001634343,0.002091715,0.003010511,0.003870832,0.003816245,0.02813021],"category_scores_gemma":[0.04692332,0.001054803,0.001710745,0.0008716488,0.001034281,0.002468697,0.006157951,0.004354811,0.02226136],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001474697,"about_ca_system_score_gemma":0.003814907,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006285737,"about_ca_topic_score_gemma":0.009333034,"domain_scores_codex":[0.9868087,0.007413812,0.0007106037,0.0021363,0.002164336,0.0007662683],"domain_scores_gemma":[0.9550875,0.02257548,0.0009911939,0.007207714,0.008852476,0.005285581],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0065567,0.001699226,0.005755154,0.002523391,0.0005457092,0.002356378,0.004303536,0.01806914,0.07032323,0.002658817,0.5453212,0.3398874],"study_design_scores_gemma":[0.004916653,0.005508635,0.02052306,0.0005960491,0.0003977232,0.004499616,0.004654307,0.4188676,0.1299978,0.01707913,0.3916197,0.001339663],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1642413,0.002710595,0.5089881,0.01117463,0.0106547,0.005892971,0.07517458,0.2004358,0.02072724],"genre_scores_gemma":[0.3484034,0.0005798835,0.4576509,0.002635899,0.002285383,0.003897988,0.1397307,0.01574385,0.02907189],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02813021,"threshold_uncertainty_score":0.09410489,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1173792312521115,"score_gpt":0.3526829460646227,"score_spread":0.2353037148125112,"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."}}