{"id":"W1966802006","doi":"10.1103/physrevstper.10.020105","title":"Assessing the quality of a student-generated question repository","year":2014,"lang":"en","type":"article","venue":"Physical Review Special Topics - Physics Education Research","topic":"Science Education and Pedagogy","field":"Social Sciences","cited_by":51,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"University of Edinburgh; Joint Information Systems Committee","keywords":"Quality (philosophy); Summative assessment; Subject (documents); Work (physics); Quality assessment; Contrast (vision)","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":[0.05560659,0.0006450468,0.001150004,0.007519387,0.0008147656,0.004909476,0.001637728,0.001429012,0.001643019],"category_scores_gemma":[0.3533013,0.0004732056,0.0008326941,0.004196597,0.001108887,0.003110837,0.003121269,0.001055922,0.0006701177],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001255889,"about_ca_system_score_gemma":0.001829044,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001047203,"about_ca_topic_score_gemma":0.00177087,"domain_scores_codex":[0.9490711,0.02369313,0.005819892,0.00349842,0.01689293,0.001024508],"domain_scores_gemma":[0.3770217,0.4676853,0.05655294,0.02760557,0.06722413,0.003910292],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001345748,0.001214007,0.5301958,0.002250552,0.0005916061,0.000494498,0.03461925,0.004035186,0.02068803,0.001056063,0.002558783,0.4009505],"study_design_scores_gemma":[0.0001659071,0.005686433,0.878934,0.001189949,0.0006786109,0.002065292,0.0172024,0.02156032,0.04607014,0.003064524,0.02293466,0.0004478056],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9722048,0.000882256,0.02158893,0.0003541287,0.00004351963,0.0004889771,0.000572963,0.0009177087,0.002946643],"genre_scores_gemma":[0.971701,0.0003350646,0.02474918,0.0001069744,0.00003470513,0.0002297789,0.001328047,0.0001865761,0.001328747],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05560659,"threshold_uncertainty_score":0.2940792,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3622447747359495,"score_gpt":0.6512391505911971,"score_spread":0.2889943758552476,"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."}}