{"id":"W4320558549","doi":"10.48550/arxiv.2302.04983","title":"CREDENCE: Counterfactual Explanations for Document Ranking","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Advanced Text Analysis Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University; York University; University of Waterloo","funders":"","keywords":"Credence; Counterfactual thinking; Ranking (information retrieval); Information retrieval; Computer science; Attribution; Machine learning; Psychology; Social psychology","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":[0.01121289,0.001396388,0.001024138,0.003978436,0.001587036,0.004711864,0.003094997,0.003334721,0.01829967],"category_scores_gemma":[0.07876221,0.0008031069,0.002430128,0.002477435,0.003533082,0.00737447,0.004462868,0.004353998,0.001633052],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001975172,"about_ca_system_score_gemma":0.001788618,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002412322,"about_ca_topic_score_gemma":0.002835365,"domain_scores_codex":[0.991839,0.005114394,0.0003339185,0.0008128894,0.001604969,0.000294921],"domain_scores_gemma":[0.9289482,0.06249204,0.002321739,0.004397428,0.001194887,0.0006456078],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0004578817,0.0001655347,0.003078077,0.0003996239,0.0001897737,0.0007541025,0.0009846159,0.1131754,0.001498533,0.7856458,0.01515488,0.07849586],"study_design_scores_gemma":[0.0000539546,0.00002685922,0.0002490732,0.00005060673,0.00002945945,0.0001284726,0.00005929439,0.3356895,0.001088797,0.655733,0.006854111,0.00003686451],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.00778035,0.0003708972,0.9823199,0.002201426,0.0001597604,0.0001003759,0.0007966839,0.002278339,0.003992277],"genre_scores_gemma":[0.4559079,0.0007964464,0.5329676,0.001038171,0.0007256979,0.0007213715,0.002110749,0.0008645353,0.004867556],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01829967,"threshold_uncertainty_score":0.0612185,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1383141408168375,"score_gpt":0.2483798991809904,"score_spread":0.1100657583641529,"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."}}