{"id":"W4361770419","doi":"10.2139/ssrn.4393234","title":"A System for the Improvement of Educational Assessment Using Intelligent Conversational Agents","year":2023,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Innovation in Digital Healthcare Systems","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":false,"ca_institutions":"Lakehead University; McMaster University; Ontario Tech University","funders":"","keywords":"Computer science; Human–computer interaction; Natural language processing; Multimedia","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":[0.005525781,0.0001248074,0.000200027,0.0002170025,0.001066394,0.00001583223,0.0002697867,0.00009195184,0.00007424304],"category_scores_gemma":[0.0001672191,0.00009106456,0.0001116914,0.0004864605,0.00004001234,0.0001190555,0.00006259739,0.001048469,0.00005357912],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.005456493,"about_ca_system_score_gemma":0.01368451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002664836,"about_ca_topic_score_gemma":0.0001457495,"domain_scores_codex":[0.9965345,0.0001760876,0.001085186,0.0001652074,0.0005283529,0.001510653],"domain_scores_gemma":[0.996942,0.001032031,0.0008396969,0.0001943033,0.0009328261,0.00005917298],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00005596639,0.00004185881,0.010232,0.0005216862,0.0002796188,1.658569e-7,0.00073313,0.0003036676,0.0003711625,0.9816493,0.00218247,0.00362896],"study_design_scores_gemma":[0.007241984,0.001872115,0.0462351,0.002924207,0.0002957412,0.0002342168,0.4476949,0.09047423,0.0004050616,0.361261,0.0403318,0.001029604],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7845652,0.0008501915,0.1525301,0.02863601,0.01890675,0.008641368,0.0002544806,0.0001602129,0.005455717],"genre_scores_gemma":[0.9954346,0.0001013651,0.0001146659,0.0002163852,0.0009612672,0.0002712846,0.00004600577,0.0000277579,0.002826608],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6203883,"threshold_uncertainty_score":0.9983614,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1010180960636076,"score_gpt":0.4730243514294722,"score_spread":0.3720062553658646,"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."}}