{"id":"W4403922859","doi":"10.1145/3686215.3688382","title":"Detecting when Users Disagree with Generated Captions","year":2024,"lang":"en","type":"article","venue":"","topic":"Video Analysis and Summarization","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Universitas Brawijaya","keywords":"Computer science; Natural language processing","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":[0.000102814,0.00005826023,0.00005698609,0.00008854127,0.0001009116,0.0004573548,0.0001727835,0.0000196966,0.00005832214],"category_scores_gemma":[0.000008227717,0.00003859527,0.00003038979,0.0005309177,0.000009753315,0.0004288222,0.00004348731,0.00005150884,0.00005707666],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002105303,"about_ca_system_score_gemma":0.00003691631,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001511861,"about_ca_topic_score_gemma":0.0007222897,"domain_scores_codex":[0.999438,0.00002133382,0.00009184709,0.0002173146,0.0001263479,0.0001051437],"domain_scores_gemma":[0.9996851,0.00002465414,0.00001206306,0.0001985788,0.00004091923,0.00003866144],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00000618723,0.00008351458,0.01422004,0.00005901529,0.000552845,0.000152885,0.007032095,0.03426663,0.02168652,0.5072945,0.01274317,0.4019026],"study_design_scores_gemma":[0.00004050292,0.00002195114,0.0004150493,0.0000138961,0.00001559782,0.000006502748,0.00004774761,0.9951543,0.001193529,0.0009486763,0.002050182,0.00009209642],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02073861,0.00006895378,0.974459,0.001042161,0.00008999644,0.00003406575,3.680296e-7,0.0002932746,0.003273577],"genre_scores_gemma":[0.9579021,0.000004215215,0.03907223,0.0001582374,0.00004087831,0.000004551828,0.000003517522,0.000005273092,0.002809037],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9608876,"threshold_uncertainty_score":0.4410283,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01576681960141888,"score_gpt":0.2208626449127426,"score_spread":0.2050958253113237,"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."}}