{"id":"W4411990669","doi":"10.5430/wjel.v15n6p385","title":"Revolutionizing Assessment: Leveraging ChatGPT for Automated Item Generation: An AI Driven Exploratory Study with EFL Teachers","year":2025,"lang":"en","type":"article","venue":"World Journal of English Language","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Exploratory research; Mathematics education; Artificial intelligence; Psychology; Sociology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0139503,0.0004719185,0.0006496605,0.00122534,0.002188701,0.002987515,0.001402872,0.001017006,0.001445243],"category_scores_gemma":[0.04163693,0.000635448,0.0003323985,0.0007802957,0.001962375,0.002325462,0.0023304,0.001540558,0.0006203369],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001537504,"about_ca_system_score_gemma":0.002033322,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003390626,"about_ca_topic_score_gemma":0.01033519,"domain_scores_codex":[0.9916357,0.006131756,0.0003532632,0.0005769834,0.0006959029,0.0006065207],"domain_scores_gemma":[0.9653168,0.02712717,0.001682487,0.001560189,0.003081791,0.001231482],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0002229682,0.004710056,0.06370448,0.0005119006,0.00002371248,0.002294606,0.8276635,0.0005453961,0.009329337,0.0007272169,0.0009121784,0.08935468],"study_design_scores_gemma":[0.0001817403,0.004266409,0.09400713,0.0005107246,0.0000682356,0.00219973,0.8583004,0.007509911,0.008634449,0.001592539,0.02256195,0.0001668411],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9960917,0.00004253063,0.002564724,0.0001286719,0.000004905632,0.0002725317,0.00002376317,0.00002340969,0.0008478586],"genre_scores_gemma":[0.9838856,0.0001595987,0.01321672,0.0002567526,0.00001112162,0.0005892179,0.00005489009,0.00002505499,0.00180101],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0139503,"threshold_uncertainty_score":0.07377714,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07488605100019907,"score_gpt":0.4139629723887305,"score_spread":0.3390769213885314,"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."}}