{"id":"W4416034664","doi":"10.18653/v1/2025.findings-emnlp.481","title":"Hard Negatives, Hard Lessons: Revisiting Training Data Quality for Robust Information Retrieval with LLMs","year":2025,"lang":"","type":"article","venue":"","topic":"Data Quality and Management","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Training (meteorology); Quality (philosophy); Data quality; Data retrieval; Training set; Information system","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06862291,0.001390786,0.002780724,0.003940195,0.00172648,0.007832369,0.005334097,0.002792265,0.003853482],"category_scores_gemma":[0.3745742,0.0009828892,0.001207566,0.002777836,0.005075014,0.0152764,0.006909281,0.006484017,0.0007622024],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003429678,"about_ca_system_score_gemma":0.005927115,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007857079,"about_ca_topic_score_gemma":0.008101701,"domain_scores_codex":[0.9504352,0.0320789,0.004447408,0.004503147,0.00718355,0.001351858],"domain_scores_gemma":[0.6626006,0.2671399,0.01070876,0.03248243,0.02490952,0.002158794],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.003462605,0.0008601464,0.03984739,0.002890597,0.001027297,0.0002701836,0.00236048,0.1072129,0.00778361,0.05198139,0.02001166,0.7622918],"study_design_scores_gemma":[0.0003036904,0.0007905283,0.008094249,0.0006143602,0.0003601563,0.0002029524,0.0007010044,0.83255,0.01153725,0.1393092,0.005398279,0.0001382029],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08060459,0.002973178,0.893328,0.0136658,0.000385778,0.0005567698,0.001122272,0.003472916,0.003890724],"genre_scores_gemma":[0.7602782,0.000462948,0.234287,0.001907872,0.0004807007,0.0002789173,0.0009942782,0.000388814,0.0009213655],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.06862291,"threshold_uncertainty_score":0.3629169,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5712677946581087,"score_gpt":0.4860539359648122,"score_spread":0.08521385869329656,"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."}}