{"id":"W6910481555","doi":"10.48448/h065-rg78","title":"Demoting the Lead Bias in News Summarization via Alternating Adversarial Learning","year":2021,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia Hospital","funders":"","keywords":"Automatic summarization; Generality; Focus (optics); Lead (geology); Adversarial system; Limiting; Media bias; Categorization","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003285916,0.0004543227,0.0004457457,0.001292812,0.0005869177,0.0004899302,0.001418812,0.0002585834,0.001105331],"category_scores_gemma":[0.003407856,0.0003739974,0.00008770633,0.003826168,0.0008902145,0.0003224624,0.0006710546,0.001211204,0.0006484977],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006011518,"about_ca_system_score_gemma":0.0009742623,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01025082,"about_ca_topic_score_gemma":0.02665835,"domain_scores_codex":[0.9952508,0.0004946332,0.0006583135,0.001142943,0.001522482,0.0009308871],"domain_scores_gemma":[0.9977438,0.0003459036,0.0009796283,0.0006267677,0.0001845174,0.0001193596],"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":[0.00008679119,0.0006912403,0.2213255,0.0004564381,0.0002771049,0.0003564119,0.008222447,0.2286619,0.1069606,0.006721352,0.03681846,0.3894217],"study_design_scores_gemma":[0.001316745,0.00007240761,0.0007093422,0.001463901,0.0000602081,0.00004749894,0.002458321,0.8949889,0.0007911797,0.0002393243,0.09677137,0.001080827],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.01142534,0.001119426,0.05072715,0.0009137452,0.00535966,0.001912351,0.00003638118,0.001079749,0.9274262],"genre_scores_gemma":[0.531477,0.0002088868,0.04194684,0.0008736423,0.007212937,0.00006419192,0.0009683741,0.003122711,0.4141254],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.6663269,"threshold_uncertainty_score":0.9998712,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05508924032628736,"score_gpt":0.3181122020825184,"score_spread":0.2630229617562311,"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."}}