{"id":"W2483081267","doi":"10.4018/978-1-4666-9441-5.ch022","title":"Using Automated Procedures to Generate Test Items That Measure Junior High Science Achievement","year":2015,"lang":"en","type":"book-chapter","venue":"Advances in higher education and professional development book series","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Section (typography); Computer science; Test (biology); Quality (philosophy); Measure (data warehouse); Curriculum; Mathematics education; Data mining; Psychology; Pedagogy","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":[0.004415728,0.001763351,0.0008559133,0.003189333,0.0005521417,0.00206675,0.001926672,0.0009971443,0.01332809],"category_scores_gemma":[0.03857334,0.000655867,0.0007799534,0.002745111,0.0005561088,0.001351571,0.001109307,0.001125369,0.01106768],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006390017,"about_ca_system_score_gemma":0.00149365,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002035091,"about_ca_topic_score_gemma":0.003061841,"domain_scores_codex":[0.9948947,0.001961617,0.0003975533,0.0008215328,0.001824212,0.0001003365],"domain_scores_gemma":[0.9676034,0.02241195,0.001574591,0.003264848,0.004905937,0.0002391892],"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":[0.000135128,0.0004527915,0.004901004,0.0005646955,0.00007367795,0.0001549201,0.0007619932,0.00641899,0.01432269,0.003842625,0.01939088,0.9489806],"study_design_scores_gemma":[0.001011532,0.002405006,0.07151505,0.001335724,0.0003910311,0.004229821,0.001966518,0.384424,0.2172794,0.04847596,0.2661841,0.0007818478],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02069064,0.000261104,0.9494643,0.0001874688,0.0002114642,0.00252927,0.001887508,0.01438689,0.01038141],"genre_scores_gemma":[0.02770072,0.000167137,0.9607342,0.0001012387,0.00004186965,0.002423669,0.002243881,0.0009785736,0.005608634],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01332809,"threshold_uncertainty_score":0.04458696,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06464199464959075,"score_gpt":0.3595467430617855,"score_spread":0.2949047484121948,"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."}}