{"id":"W7015576738","doi":"","title":"TagDebias: Entity and Concept Typing for Social Bias Mitigation in Pretrained Language Models","year":2024,"lang":"fr","type":"other","venue":"PolyPublie (École Polytechnique de Montréal)","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Mitacs","keywords":"Corpus linguistics; Personal pronoun","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":{"n_in":0,"stratum":"french","weight":1554.46666666667,"opus":{"tier":"OUT","genre":"empirical","about_ca":false,"confidence":"medium","reason":"Thesis proposing a tagging approach to mitigate social bias in pretrained language models; AI-fairness method development, where bias refers to model bias, not research bias."},"gpt":{"tier":"OUT","genre":"empirical","about_ca":false,"confidence":"high","reason":"It develops a method for mitigating bias in language models, not bias or practice in research."},"grok":{"tier":"OUT","genre":"empirical","about_ca":false,"confidence":"high","reason":"NLP method to mitigate gender bias in pretrained language models; AI fairness, not study of research practice."}},"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005500889,0.001579529,0.001014793,0.001796648,0.0009515996,0.003182423,0.002072681,0.001768289,0.00574814],"category_scores_gemma":[0.02130455,0.001041939,0.001568415,0.001305884,0.0009518294,0.007721701,0.003492099,0.003511498,0.002760765],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00141888,"about_ca_system_score_gemma":0.002999976,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01053396,"about_ca_topic_score_gemma":0.02314225,"domain_scores_codex":[0.997399,0.001027821,0.0002509676,0.0005973972,0.0005670369,0.0001578065],"domain_scores_gemma":[0.9897626,0.006203706,0.0004456091,0.002004657,0.001325137,0.0002582854],"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":[0.00170605,0.0004880379,0.0334466,0.001333127,0.0006280093,0.0006136926,0.004834839,0.1645569,0.01947129,0.03190433,0.02505402,0.7159631],"study_design_scores_gemma":[0.00008375153,0.0001645842,0.002711829,0.0001873516,0.0001496942,0.0002006847,0.00069543,0.9245377,0.01321818,0.03326831,0.02469683,0.00008553494],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.07784127,0.001073526,0.8936594,0.001019172,0.0003839662,0.0003283817,0.002693402,0.02040371,0.002597276],"genre_scores_gemma":[0.4269634,0.0006835853,0.552446,0.0005636103,0.0001356345,0.0006242954,0.006661778,0.002639333,0.009282338],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01053396,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02131562394811607,"score_gpt":0.2668336012770453,"score_spread":0.2455179773289292,"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."}}