{"id":"W3175609992","doi":"10.21428/594757db.ae6ae665","title":"Direct Answer Threshold Optimization in Dialogue Systems","year":2021,"lang":"en","type":"article","venue":"","topic":"Speech and dialogue systems","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Bank of Canada","funders":"","keywords":"Computer science; Set (abstract data type); Point (geometry); Focus (optics); Task (project management); Information retrieval; Machine learning; Artificial intelligence; Mathematics","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":[0.0002547123,0.00009244867,0.0001716118,0.00009037709,0.00003472904,0.0002288308,0.0003005744,0.00006818791,0.00002367409],"category_scores_gemma":[0.00005440196,0.00008164665,0.00003835139,0.0006133259,0.00001020264,0.0003529623,0.0001042532,0.00005516012,0.00007433025],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004217202,"about_ca_system_score_gemma":0.00009784207,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003347158,"about_ca_topic_score_gemma":0.0002256565,"domain_scores_codex":[0.9989336,0.00008492311,0.0002274111,0.0003345411,0.0001943382,0.0002252416],"domain_scores_gemma":[0.9992916,0.00005668008,0.00003691564,0.0004630924,0.00008066885,0.00007103861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001669465,0.0004087526,0.03144368,0.0001211414,0.0000497672,0.0009427122,0.00140327,0.6685091,0.002779703,0.2713512,0.02043258,0.00254149],"study_design_scores_gemma":[0.001020172,0.0000548679,0.002804736,0.00009853946,0.000004271224,0.00008876077,0.0001702349,0.9799051,0.008660079,0.0004859946,0.006207824,0.0004994125],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002973034,0.0009446155,0.7845942,0.0003476319,0.002934046,0.0002209743,0.000003007736,0.0002713991,0.207711],"genre_scores_gemma":[0.9857115,0.00002771797,0.01200156,0.0002120612,0.0001555489,0.00002601502,0.00001919515,0.000007769129,0.001838707],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9827384,"threshold_uncertainty_score":0.3329454,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01697210482856354,"score_gpt":0.2254107716638141,"score_spread":0.2084386668352506,"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."}}