{"id":"W4410086952","doi":"10.1109/wi-iat62293.2024.00028","title":"HouseOfTheDragonQA: Open-Domain Long-form Context-Aware QA Pairs for TV Series","year":2024,"lang":"en","type":"article","venue":"","topic":"Multimedia Communication and Technology","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"Series (stratigraphy); Computer science; Context (archaeology); Domain (mathematical analysis); Open domain; Information retrieval; Mathematics; History","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007445142,0.0000983056,0.0001585746,0.00008087842,0.0005325818,0.0002781964,0.001020473,0.0001686688,0.001649128],"category_scores_gemma":[0.0001835764,0.00008309964,0.00007206734,0.0002579133,0.0004329407,0.0004097968,0.0002564199,0.0001387781,0.0002001691],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008067284,"about_ca_system_score_gemma":0.0002348519,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001735471,"about_ca_topic_score_gemma":0.04474698,"domain_scores_codex":[0.9990925,0.00008362735,0.0001856805,0.0002166363,0.0001328554,0.0002886687],"domain_scores_gemma":[0.9990719,0.0003706022,0.00003395365,0.0003669625,0.00007449527,0.0000821126],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001517615,0.00003523915,0.0003703237,0.00001868853,0.00003047771,0.000003289384,0.00809053,1.060509e-7,0.00002005582,0.8837574,0.04079963,0.06685906],"study_design_scores_gemma":[0.000244341,0.00004831631,0.0001104342,0.00002335719,0.000007429769,0.00000207055,0.01899354,0.0001273526,0.0002687573,0.04075403,0.9392833,0.0001371244],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.02715028,0.002674757,0.03828755,0.3651129,0.002862092,0.005090082,0.0001242571,0.005305854,0.5533922],"genre_scores_gemma":[0.9618877,0.0003228849,0.003998033,0.000774527,0.00009620474,0.0002787371,0.00001805626,0.00002185652,0.03260204],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9347374,"threshold_uncertainty_score":0.9992635,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05709954398164591,"score_gpt":0.3721838186245334,"score_spread":0.3150842746428875,"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."}}