{"id":"W4414360309","doi":"10.24963/ijcai.2025/643","title":"DUQ: Dual Uncertainty Quantification for Text-Video Retrieval","year":2025,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Similarity (geometry); Granularity; Construct (python library); Benchmark (surveying); Dual (grammatical number); Feature (linguistics); Uncertainty reduction theory; Video retrieval; Trustworthiness","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":[0.002807529,0.001120318,0.001354886,0.002003776,0.0005519607,0.001905708,0.002581698,0.001294139,0.002253157],"category_scores_gemma":[0.01158444,0.0005089413,0.0009366892,0.001820101,0.00108046,0.005036054,0.003399248,0.001685378,0.0006596001],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002124694,"about_ca_system_score_gemma":0.001320347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00520453,"about_ca_topic_score_gemma":0.003572399,"domain_scores_codex":[0.9973015,0.0005982005,0.0001876872,0.0005411568,0.001187942,0.0001835219],"domain_scores_gemma":[0.9969541,0.001613408,0.0003787745,0.0004048494,0.0005395642,0.0001093531],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0004408348,0.000191828,0.002012792,0.0004201943,0.0001522547,0.0002040016,0.0003511184,0.4270241,0.01862377,0.06566545,0.008157915,0.4767557],"study_design_scores_gemma":[0.00001045151,0.00004672818,0.0002171139,0.00001082027,0.00001633004,0.0000524098,0.00002022778,0.9787176,0.00289186,0.01627554,0.001719965,0.00002094824],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.006019543,0.0006562141,0.9913275,0.000179834,0.00002871225,0.00007168399,0.000156839,0.0006251528,0.0009345568],"genre_scores_gemma":[0.6072689,0.0009191922,0.3853847,0.0004962005,0.0002445872,0.000402934,0.001021081,0.000262287,0.003999994],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.00520453,"threshold_uncertainty_score":0.01541585,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01876358955548414,"score_gpt":0.3142187101013578,"score_spread":0.2954551205458736,"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."}}