{"id":"W3206421219","doi":"10.18280/ts.400214","title":"Statistical Evaluation of Video Summarization Models from an Empirical Perspective","year":2023,"lang":"en","type":"article","venue":"Traitement du signal","topic":"Video Analysis and Summarization","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Automatic summarization; Perspective (graphical); Computer science; Empirical research; Artificial intelligence; Statistics; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01939141,0.001389072,0.001084204,0.004165366,0.0004371748,0.00178583,0.001510477,0.001221141,0.001355577],"category_scores_gemma":[0.08010544,0.0002847989,0.0009472914,0.002859618,0.000820527,0.002924073,0.0008225957,0.001373612,0.0004778151],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002011802,"about_ca_system_score_gemma":0.0009927645,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004911671,"about_ca_topic_score_gemma":0.003524599,"domain_scores_codex":[0.9930931,0.0034469,0.0006376525,0.00111899,0.00147697,0.0002264184],"domain_scores_gemma":[0.9047916,0.07681961,0.00545244,0.004363589,0.007876844,0.0006958596],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001482209,0.0008055584,0.05475869,0.001283841,0.001118924,0.0001825727,0.0004656811,0.6379908,0.002969228,0.006359837,0.01240215,0.2801804],"study_design_scores_gemma":[0.00003865752,0.0007712357,0.01443359,0.00009902848,0.0001426028,0.0001005379,0.000240851,0.9774848,0.001755332,0.003262349,0.001625299,0.00004581859],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5824028,0.01163189,0.3871335,0.002144417,0.0004952104,0.0008246815,0.005858509,0.003136178,0.006372836],"genre_scores_gemma":[0.9390324,0.001535863,0.04873386,0.0001691322,0.0002770523,0.0004356442,0.008659173,0.0001720213,0.0009847099],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01939141,"threshold_uncertainty_score":0.1025528,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07721719914288039,"score_gpt":0.3531169046251268,"score_spread":0.2758997054822465,"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."}}