{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004190428,0.001672194,0.001500536,0.002912894,0.001141185,0.002818675,0.002143109,0.002181212,0.009513513],"category_scores_gemma":[0.01623295,0.0007581313,0.000679932,0.0009626477,0.0004140295,0.002843371,0.003096205,0.001860693,0.007186683],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007381025,"about_ca_system_score_gemma":0.001787396,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003647214,"about_ca_topic_score_gemma":0.002089323,"domain_scores_codex":[0.99597,0.001479292,0.0005451368,0.0009128535,0.0009188312,0.0001739452],"domain_scores_gemma":[0.9898559,0.005638595,0.000591471,0.001131234,0.001935907,0.0008469551],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003725572,0.001678378,0.01068347,0.0008415454,0.0003388061,0.0006343673,0.002207558,0.00522369,0.04918694,0.004363972,0.03491862,0.886197],"study_design_scores_gemma":[0.001012953,0.001532617,0.01485369,0.0004864685,0.0009833834,0.001144856,0.0009081843,0.717163,0.1187079,0.01230625,0.1302564,0.0006443226],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04460644,0.0007402378,0.6799427,0.0006873315,0.0004067346,0.001596516,0.001935407,0.2626262,0.007458445],"genre_scores_gemma":[0.2805587,0.0003688836,0.6965263,0.0007019326,0.0002780076,0.002130364,0.003724706,0.002762978,0.01294813],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.009513513,"threshold_uncertainty_score":0.0318259,"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."}}