{"id":"W4389925007","doi":"10.21449/ijate.1359348","title":"Automatic item generation for non-verbal reasoning items","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Verbal reasoning; Computer science; Test (biology); Item analysis; Cognition; Item response theory; Homogeneous; Subject-matter expert; Natural language processing; Artificial intelligence; Subject matter; Psychology; Cognitive psychology; Psychometrics; Pedagogy; Expert system; Mathematics; Developmental psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.002158554,0.0001006447,0.0001609758,0.0005695026,0.0001625959,0.0003230543,0.0004285853,0.00008128827,0.0001928821],"category_scores_gemma":[0.000653055,0.0001051719,0.0001086955,0.000389091,0.00004130116,0.001060117,0.00002348468,0.0001596002,0.00001493514],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009298055,"about_ca_system_score_gemma":0.003643859,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001984725,"about_ca_topic_score_gemma":0.0002967454,"domain_scores_codex":[0.9980417,0.0001447441,0.0006226869,0.0001520298,0.000829571,0.000209198],"domain_scores_gemma":[0.9976373,0.0006620656,0.0005375269,0.00008477119,0.0009941524,0.000084147],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00005804947,0.001401331,0.2410517,0.00005184151,0.0002689316,0.00001221194,0.02523632,0.00127119,0.004673972,0.1617295,0.09701511,0.4672298],"study_design_scores_gemma":[0.002073143,0.0002857038,0.6352672,0.0006652814,0.00008064404,0.00003102356,0.05633244,0.05547003,0.0003034543,0.01978774,0.229109,0.0005943266],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9672849,0.00004639044,0.001869205,0.01173054,0.01211992,0.0003714281,0.000009762377,0.00002946155,0.006538349],"genre_scores_gemma":[0.984994,0.0002382532,0.008437909,0.0002445746,0.004483025,0.00009518263,0.0001776082,0.00001352426,0.001315915],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4666355,"threshold_uncertainty_score":0.6464049,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08817372903914804,"score_gpt":0.4972479358557301,"score_spread":0.4090742068165821,"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."}}