{"id":"W4313410276","doi":"10.1111/emip.12537","title":"Using Active Learning Methods to Strategically Select Essays for Automated Scoring","year":2022,"lang":"en","type":"article","venue":"Educational Measurement Issues and Practice","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre for Advancing Health Outcomes; University of Alberta","funders":"","keywords":"Computer science; Artificial intelligence; Machine learning; Scalability; Active learning (machine learning); Encoder; Transformer; Database; Engineering","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":[0.01562862,0.0001633782,0.0001947711,0.0001715124,0.001185579,0.00008752313,0.0001350655,0.00004298462,0.0008784932],"category_scores_gemma":[0.007200068,0.0001748068,0.00003965682,0.0003704728,0.00002393888,0.0001820065,0.00006125948,0.0006333948,0.000008827799],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002882008,"about_ca_system_score_gemma":0.0002601426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004462705,"about_ca_topic_score_gemma":0.000001725886,"domain_scores_codex":[0.9921517,0.006381372,0.0002697522,0.0004365322,0.0004546564,0.0003059548],"domain_scores_gemma":[0.9961637,0.002719143,0.0002639127,0.0001428696,0.0006252194,0.0000851161],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.006707138,0.003153496,0.005395372,0.0001911587,0.002414115,0.00001157611,0.1569949,0.04890833,0.2272839,0.2500341,0.01912157,0.2797843],"study_design_scores_gemma":[0.0009108151,0.0011308,0.01750909,0.0000509554,0.000220843,0.000176937,0.0367869,0.003835209,0.0006428717,0.002928182,0.9352193,0.0005880861],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1618159,0.00530098,0.612847,0.05863455,0.007582868,0.003125969,0.00003469316,0.0009440279,0.149714],"genre_scores_gemma":[0.3901102,0.000003001759,0.6057253,0.000819337,0.0005210682,0.0003524213,0.00002104359,0.00004000297,0.002407686],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9160978,"threshold_uncertainty_score":0.9618885,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3365902019122547,"score_gpt":0.5652165547572977,"score_spread":0.228626352845043,"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."}}