{"id":"W2526840435","doi":"10.1109/dmiaf.2016.7574940","title":"Assessing unreliability in OTT video QoE subjective evaluations using clustering with idealized data","year":2016,"lang":"en","type":"article","venue":"","topic":"Image and Video Quality Assessment","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Telus (Canada)","funders":"","keywords":"Cluster analysis; Computer science; Quality of experience; Reliability (semiconductor); Consistency (knowledge bases); Set (abstract data type); Perspective (graphical); Point (geometry); Data set; Video quality; Subjective video quality; Data mining; Machine learning; Artificial intelligence; Computer network; Engineering; Image quality; Metric (unit); Image (mathematics); Quality of service; 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":[0.002608808,0.0001632449,0.0002301083,0.0001397941,0.0001586617,0.0004598695,0.001068709,0.00004857374,0.00004937884],"category_scores_gemma":[0.0003330803,0.0001038073,0.00002730066,0.0005561708,0.00007552318,0.005062372,0.0009837564,0.0001075001,0.00001524855],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003277025,"about_ca_system_score_gemma":0.0005195696,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001159975,"about_ca_topic_score_gemma":0.00128227,"domain_scores_codex":[0.997508,0.000481172,0.0004066341,0.0007678134,0.000474604,0.0003618112],"domain_scores_gemma":[0.9973083,0.0005940431,0.0001341224,0.001705125,0.0001821507,0.00007631717],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002615787,0.001784283,0.1727358,0.0003348557,0.000318472,0.0002060272,0.007435419,0.007675388,0.1035894,0.05687113,0.0007851122,0.6480025],"study_design_scores_gemma":[0.002608394,0.000101831,0.09190252,0.0004385707,0.00003077009,0.00003090913,0.0005648035,0.8923771,0.004460475,0.006649134,0.0002637156,0.0005717924],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09190097,0.0000153036,0.9048245,0.001530308,0.0001340741,0.0002796866,0.000004533223,0.0001081722,0.001202453],"genre_scores_gemma":[0.7550786,0.000002420735,0.2445752,0.0002351377,0.00002925249,0.00001095898,0.00000323109,0.000008445543,0.00005685686],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8847017,"threshold_uncertainty_score":0.4434532,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1813825072810651,"score_gpt":0.4451307727335275,"score_spread":0.2637482654524624,"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."}}