{"id":"W2915483812","doi":"10.1145/3287324.3293827","title":"Evaluating the Effect of Follow-up Questions in an Online Exercise","year":2019,"lang":"en","type":"article","venue":"","topic":"Online Learning and Analytics","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Class (philosophy); Sentence; Quality (philosophy); Mathematics education; Significant difference; Psychology; Computer science; Cognitive psychology; Artificial intelligence; Medicine","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":[0.00512186,0.001387365,0.0009586419,0.0005049625,0.0003497744,0.0009985018,0.001054156,0.00155132,0.005038993],"category_scores_gemma":[0.05963345,0.0003576449,0.0006477957,0.0003385042,0.000541969,0.00110688,0.0009763491,0.001378826,0.0009386306],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003186557,"about_ca_system_score_gemma":0.0004263676,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003916274,"about_ca_topic_score_gemma":0.0006002217,"domain_scores_codex":[0.9955755,0.002113167,0.0005686733,0.0007347626,0.0006499376,0.0003579072],"domain_scores_gemma":[0.8555987,0.125172,0.007885219,0.004843412,0.002908637,0.003592102],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.07303897,0.1272888,0.03510525,0.00302527,0.0007173074,0.0003838392,0.006789103,0.007201602,0.398653,0.0005400738,0.001419561,0.3458371],"study_design_scores_gemma":[0.008637077,0.3514564,0.3672187,0.0003889519,0.001447828,0.0003564087,0.001812721,0.01493328,0.2451787,0.001548213,0.006745188,0.0002766504],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9958098,0.00006313901,0.002403477,0.00006650181,0.00004061345,0.0004630424,0.00006986612,0.000110378,0.0009732014],"genre_scores_gemma":[0.9743816,0.0001348369,0.02117621,0.0001599891,0.00008321808,0.001335513,0.0002957205,0.00006838951,0.002364559],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00512186,"threshold_uncertainty_score":0.02708727,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03591639783005256,"score_gpt":0.3706196801546572,"score_spread":0.3347032823246047,"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."}}