{"id":"W4404873315","doi":"10.1177/13563890241289937","title":"Artificial intelligence and big data-driven evaluation research and practices: A systematic literature review","year":2024,"lang":"en","type":"article","venue":"Evaluation","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"École Nationale d'Administration Publique","funders":"","keywords":"Big data; Systematic review; Data science; Management science; Psychology; Political science; Engineering ethics; Computer science; MEDLINE; Engineering; Data mining","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":[],"category_scores_codex":[0.1039968,0.001287521,0.005738563,0.02146012,0.001878686,0.006941498,0.002701673,0.002413948,0.003125309],"category_scores_gemma":[0.2801399,0.001354327,0.004945094,0.0226651,0.003442044,0.006510608,0.004535247,0.002283371,0.0003768581],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01367275,"about_ca_system_score_gemma":0.0582359,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009866173,"about_ca_topic_score_gemma":0.0267892,"domain_scores_codex":[0.8903001,0.06455066,0.02685221,0.00257434,0.01457648,0.001146106],"domain_scores_gemma":[0.660991,0.2793103,0.02748351,0.004999712,0.02527837,0.001937008],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0002105867,0.0001095495,0.002160938,0.7517055,0.002460951,0.00015673,0.003300804,0.0004837354,0.0001502077,0.004510443,0.004660712,0.2300899],"study_design_scores_gemma":[0.00006700449,0.00009287777,0.001715864,0.9673581,0.003858765,0.0001785243,0.001814962,0.0001636389,0.0001326368,0.001836802,0.02273814,0.00004265861],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00204764,0.9907728,0.001627788,0.002307979,0.0001834425,0.00143141,0.0003377927,0.00002175169,0.001269467],"genre_scores_gemma":[0.04541507,0.9395728,0.008158366,0.001772647,0.0001194009,0.004344621,0.0004156144,0.00002977276,0.0001717924],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8960032,"threshold_uncertainty_score":0.549994,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7863038003097949,"score_gpt":0.6347017197247322,"score_spread":0.1516020805850626,"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."}}