{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0204903,0.0001198551,0.000251804,0.0003052576,0.0001843712,0.0002883366,0.0001009128,0.000116811,0.0001043049],"category_scores_gemma":[0.01463605,0.00009511757,0.00002030073,0.0009663841,0.00009481312,0.0005338216,0.00007143068,0.000413344,0.0001020612],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002062029,"about_ca_system_score_gemma":0.0008300192,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000676982,"about_ca_topic_score_gemma":0.0001036864,"domain_scores_codex":[0.9960427,0.001110643,0.0006599742,0.0005814983,0.001402827,0.0002024096],"domain_scores_gemma":[0.9961234,0.001118078,0.0001843197,0.0006118819,0.001841484,0.0001208684],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00004357381,0.0000900688,0.0001330798,0.1223612,0.00005564388,0.000009223583,0.004534952,0.000009416062,0.000246992,0.00135644,0.00220381,0.8689556],"study_design_scores_gemma":[0.00002414569,0.0004920365,0.0003156583,0.1957133,0.001403537,0.0002951445,0.003087177,0.7681053,0.0005056121,0.02890847,0.0009605563,0.0001890248],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.09212957,0.8570651,0.001789936,0.03568389,0.001933418,0.01049172,0.00002524846,0.0001108331,0.0007703073],"genre_scores_gemma":[0.9273803,0.06939211,0.0004575901,0.0003185905,0.001188941,0.0005365992,0.0006111516,0.00002006158,0.00009470512],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8687665,"threshold_uncertainty_score":0.9936641,"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."}}