{"id":"W3095278952","doi":"10.2196/21620","title":"Pupillary Responses for Cognitive Load Measurement to Classify Difficulty Levels in an Educational Video Game: Empirical Study","year":2020,"lang":"en","type":"article","venue":"JMIR Serious Games","topic":"Visual and Cognitive Learning Processes","field":"Psychology","cited_by":35,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Pupillary response; Wilcoxon signed-rank test; Pupillometry; Computer science; Pupil size; Cognitive load; Video game; Cognition; Artificial intelligence; Machine learning; Psychology; Statistics; Mathematics; Pupil; Multimedia; Mann–Whitney U test","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0003997776,0.0002852441,0.0003689216,0.0001217623,0.0001096649,0.00008630754,0.0002463894,0.00012162,0.0004442308],"category_scores_gemma":[0.001426475,0.0002597916,0.00006969822,0.0004746518,0.00007018331,0.0001275072,0.00007912265,0.0002883166,0.0003100726],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001276544,"about_ca_system_score_gemma":0.0004136319,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000277464,"about_ca_topic_score_gemma":0.000349958,"domain_scores_codex":[0.9973085,0.0004972378,0.0003942097,0.0008172662,0.000536461,0.0004463267],"domain_scores_gemma":[0.9982419,0.0004117139,0.0001094429,0.0001647946,0.0007768855,0.0002952763],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0164489,0.01097344,0.5102007,0.0002071453,0.0004659336,0.0001291782,0.2336361,0.00001624852,0.005281729,0.0001396935,0.02257954,0.1999215],"study_design_scores_gemma":[0.002384636,0.00484039,0.9552492,0.00007263863,0.00004047556,0.000007326278,0.02386056,0.00001291503,0.0001904841,0.0001036674,0.01287316,0.0003645313],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9869996,0.0002365803,0.0001766441,0.009262209,0.0002298827,0.002211069,0.00009952214,0.0001104769,0.0006740222],"genre_scores_gemma":[0.9904067,0.000001601177,0.00005961938,0.005548059,0.000388022,0.002544506,0.00002454514,0.00004505467,0.0009818793],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4450486,"threshold_uncertainty_score":0.9999855,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1991487285468198,"score_gpt":0.4680045419754571,"score_spread":0.2688558134286373,"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."}}