{"id":"W4413367821","doi":"10.1002/uro2.70028","title":"Assessing the utility of a natural language processing model in answering common urological questions","year":2025,"lang":"en","type":"article","venue":"UroPrecision","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Question answering; Computer science; Natural language processing; Natural (archaeology); Information retrieval; History","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.03304775,0.0009457637,0.0008249537,0.002353294,0.0008548158,0.003549234,0.001364708,0.001936234,0.001753906],"category_scores_gemma":[0.1803856,0.0004287687,0.001429522,0.00136373,0.0009546034,0.002904896,0.002040642,0.001466891,0.0006352191],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002506993,"about_ca_system_score_gemma":0.002679957,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01070911,"about_ca_topic_score_gemma":0.009050538,"domain_scores_codex":[0.9703916,0.02082863,0.0019361,0.002610207,0.003742866,0.0004906484],"domain_scores_gemma":[0.7305365,0.2413839,0.006598122,0.004489246,0.01567062,0.001321607],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.009130185,0.004243771,0.5133842,0.002973377,0.001364096,0.0008609004,0.01410119,0.0975681,0.01123219,0.001854612,0.007592157,0.3356952],"study_design_scores_gemma":[0.0006142618,0.004963819,0.1308891,0.000620562,0.0009789345,0.0006603949,0.004943531,0.8390262,0.008338677,0.003841017,0.004879921,0.0002436549],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9175476,0.0003174007,0.07138139,0.001104211,0.00008731917,0.002283501,0.001043656,0.001623666,0.004611304],"genre_scores_gemma":[0.9017906,0.0001181614,0.09464885,0.0003638866,0.00004362188,0.001193276,0.001339487,0.00007790217,0.0004240657],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03304775,"threshold_uncertainty_score":0.1747753,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1346930753895011,"score_gpt":0.4995114290061578,"score_spread":0.3648183536166567,"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."}}