{"id":"W6929142849","doi":"10.48448/80hd-b327","title":"Realistic Conversational Question Answering with Answer Selection based on Calibrated Confidence and Uncertainty Measurement","year":2023,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Immunodeficiency and Autoimmune Disorders","field":"Immunology and Microbiology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Conversation; Question answering; Filter (signal processing); Selection (genetic algorithm); Smoothing; Paragraph; Model selection","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":[0.004987262,0.002268867,0.001123563,0.001743653,0.000979612,0.002138994,0.003736749,0.002998688,0.007321867],"category_scores_gemma":[0.01976711,0.0006308354,0.001677583,0.001056187,0.00084813,0.004001564,0.002902335,0.002599011,0.004579706],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001900661,"about_ca_system_score_gemma":0.001978454,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01809564,"about_ca_topic_score_gemma":0.0224852,"domain_scores_codex":[0.9945856,0.002294068,0.0003056348,0.001831651,0.0007131628,0.0002699663],"domain_scores_gemma":[0.9935847,0.003679388,0.000244995,0.0008570859,0.00139697,0.000237013],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002255414,0.0008335151,0.02105646,0.002905746,0.0006593245,0.0008662716,0.00269678,0.1242524,0.02843793,0.01936782,0.2115564,0.5851119],"study_design_scores_gemma":[0.0001223613,0.0001868207,0.003975509,0.0001223754,0.0001316285,0.0004076591,0.0004713716,0.9237745,0.01173181,0.02620336,0.03278019,0.00009232882],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.100842,0.006700976,0.7793902,0.004680599,0.0006438286,0.001324052,0.03374015,0.05373474,0.01894343],"genre_scores_gemma":[0.5429507,0.0008133182,0.3764192,0.002151691,0.0003662745,0.0009998373,0.06480764,0.001174438,0.01031703],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01809564,"threshold_uncertainty_score":0.03598064,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01603426615987573,"score_gpt":0.2454133938802137,"score_spread":0.2293791277203379,"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."}}