{"id":"W4416673031","doi":"10.1111/medu.70101","title":"A validity evaluation of lexicon‐based sentiment analysis of medical students' clinical performance from in‐training evaluation reports","year":2025,"lang":"en","type":"article","venue":"Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"Cumming School of Medicine, University of Calgary","keywords":"Sentiment analysis; Narrative; MEDLINE; Clinical judgment; Work (physics); Educational measurement","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":[0.06646608,0.0005661845,0.0007828679,0.004042729,0.000971273,0.002608702,0.001165111,0.0007820291,0.001483139],"category_scores_gemma":[0.1670639,0.0003424222,0.001648197,0.002467162,0.001708451,0.00203979,0.003466684,0.0006320953,0.0005794642],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001315005,"about_ca_system_score_gemma":0.001263865,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001110903,"about_ca_topic_score_gemma":0.001522221,"domain_scores_codex":[0.9569005,0.02194723,0.005796518,0.003296046,0.01122679,0.0008328418],"domain_scores_gemma":[0.7804098,0.1220538,0.03252326,0.01399512,0.04904997,0.001967994],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001972193,0.0003669835,0.9030177,0.0006341618,0.0008209168,0.000119154,0.007762461,0.001271489,0.005025858,0.0009393368,0.001700363,0.07636923],"study_design_scores_gemma":[0.0002013733,0.001678204,0.9573423,0.0003590622,0.0004810384,0.0004384029,0.004802369,0.02012728,0.007776341,0.001661896,0.004991945,0.0001397784],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9749314,0.0002819604,0.01759379,0.0002330807,0.0001209405,0.0009945307,0.0008715323,0.0001091188,0.004863604],"genre_scores_gemma":[0.9863259,0.00008188563,0.01079943,0.00008827785,0.00006057026,0.001115027,0.001083745,0.00005964173,0.000385572],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06646608,"threshold_uncertainty_score":0.3515104,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1203253672887233,"score_gpt":0.5422086184698032,"score_spread":0.4218832511810799,"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."}}