{"id":"W4379929879","doi":"10.21449/ijate.1249297","title":"Automatic item generation for online measurement and evaluation: Turkish literature items","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Turkish; Computer science; Field (mathematics); Test (biology); Item bank; Item analysis; Subject-matter expert; Subject matter; Item response theory; Data science; Artificial intelligence; Statistics; Psychometrics; Curriculum; Expert system; Psychology; Mathematics","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":[],"consensus_categories":[],"category_scores_codex":[0.01552993,0.001246164,0.001001037,0.006131188,0.0009967891,0.001400196,0.0013269,0.0006084226,0.006713116],"category_scores_gemma":[0.05633269,0.0005881862,0.001299735,0.005073262,0.0006764896,0.001537247,0.001756765,0.0007798601,0.003360545],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001402663,"about_ca_system_score_gemma":0.002453501,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001957287,"about_ca_topic_score_gemma":0.001994248,"domain_scores_codex":[0.979719,0.01119844,0.003979807,0.001340062,0.003200485,0.0005621973],"domain_scores_gemma":[0.9613953,0.01355043,0.001960424,0.004185921,0.01860139,0.0003063599],"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":[0.001337241,0.001071573,0.04157918,0.002275782,0.0001766668,0.0003749204,0.004179399,0.003309284,0.01496092,0.004148552,0.01444819,0.9121382],"study_design_scores_gemma":[0.002223251,0.008177832,0.5890786,0.003225258,0.001369474,0.002666665,0.01358815,0.1040495,0.1049058,0.0200555,0.1497973,0.0008626974],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5767459,0.001414564,0.363019,0.0009154749,0.0006681514,0.02372806,0.005143011,0.003812202,0.02455371],"genre_scores_gemma":[0.5233265,0.0005589654,0.4433104,0.0001639862,0.00008130272,0.02192322,0.006632031,0.0003631559,0.003640528],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01552993,"threshold_uncertainty_score":0.08213109,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1047808688677508,"score_gpt":0.4375316124065066,"score_spread":0.3327507435387558,"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."}}