{"id":"W4401314120","doi":"10.18260/1-2--47341","title":"Evaluating ChatGPT's Efficacy in Qualitative Analysis of Engineering Education Research","year":2024,"lang":"en","type":"article","venue":"","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"San Francisco State University; American Society for Engineering Education; National Science Foundation","keywords":"Rubric; Protocol (science); Computer science; Qualitative research; Consistency (knowledge bases); Coding (social sciences); Process (computing); Qualitative property; Medical education; Psychology; Artificial intelligence; Mathematics education; Medicine","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3817262,0.00124155,0.001424548,0.005785793,0.003835643,0.007073554,0.003115627,0.001650106,0.005200837],"category_scores_gemma":[0.619268,0.001252025,0.001255049,0.003824805,0.007596445,0.007307111,0.0144488,0.002356632,0.001321044],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006925774,"about_ca_system_score_gemma":0.005998928,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001054859,"about_ca_topic_score_gemma":0.002190078,"domain_scores_codex":[0.6274694,0.3352958,0.01197614,0.007579096,0.01491152,0.002768049],"domain_scores_gemma":[0.1568896,0.7525686,0.01665754,0.02611093,0.04554764,0.002225676],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0008803954,0.0003859395,0.01623056,0.004986816,0.0001050406,0.0004534327,0.7868919,0.002094836,0.007547956,0.008151805,0.002831114,0.1694402],"study_design_scores_gemma":[0.00107472,0.00482149,0.04917029,0.0173361,0.0004790313,0.001514017,0.7279367,0.05748789,0.03289058,0.03824518,0.06835651,0.00068749],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6309112,0.0006793172,0.3226308,0.003571552,0.000464044,0.02330842,0.001139972,0.001847223,0.01544748],"genre_scores_gemma":[0.6579269,0.0003106512,0.299167,0.0008901736,0.00008277283,0.03906919,0.000416765,0.0004374317,0.001699267],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6182737,"threshold_uncertainty_score":0.7624416,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5868279347479567,"score_gpt":0.6888672797808777,"score_spread":0.102039345032921,"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."}}