{"id":"W3164655521","doi":"10.1016/j.nepr.2021.103085","title":"Automated essay scoring (AES) of constructed responses in nursing examinations: An evaluation","year":2021,"lang":"en","type":"article","venue":"Nurse Education in Practice","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Reliability (semiconductor); Cohort; Medicine; Psychology; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01225258,0.00100904,0.0009492934,0.002668181,0.0003355321,0.00160569,0.001529803,0.001109705,0.003866516],"category_scores_gemma":[0.05832431,0.0002973461,0.00067323,0.001543518,0.0004232797,0.001334958,0.001833146,0.0007614666,0.002209711],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004390437,"about_ca_system_score_gemma":0.0007509531,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001433687,"about_ca_topic_score_gemma":0.001768803,"domain_scores_codex":[0.9837033,0.008793357,0.001291151,0.001307019,0.004619524,0.0002857989],"domain_scores_gemma":[0.8782013,0.08318675,0.004658881,0.005317804,0.02684191,0.001793329],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01120912,0.001837467,0.1300468,0.0007747242,0.0004454904,0.0002367146,0.001016861,0.004974198,0.01213999,0.000499631,0.004327023,0.8324918],"study_design_scores_gemma":[0.001903191,0.01274324,0.6915779,0.0004824776,0.00107709,0.003138941,0.002098911,0.2267882,0.04241206,0.002304203,0.01517179,0.0003020391],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9112366,0.001618758,0.07374129,0.0002504559,0.0003134367,0.001178549,0.00266712,0.003837186,0.005156712],"genre_scores_gemma":[0.887031,0.001156368,0.1003484,0.0002208204,0.0002512687,0.000690848,0.003723554,0.0003684289,0.006209174],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01225258,"threshold_uncertainty_score":0.06479865,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04273014382337106,"score_gpt":0.4885102855207469,"score_spread":0.4457801416973759,"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."}}