{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03899282,0.0005127725,0.001091521,0.0006255954,0.001284381,0.004425523,0.00422833,0.0002170267,0.0006581152],"category_scores_gemma":[0.02664221,0.0004003501,0.0001843335,0.002626767,0.0004470315,0.01312602,0.002621526,0.0004451643,0.0001816945],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001653003,"about_ca_system_score_gemma":0.001012815,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002841774,"about_ca_topic_score_gemma":0.0001615413,"domain_scores_codex":[0.9906761,0.001051321,0.003436752,0.001614366,0.002318646,0.0009027815],"domain_scores_gemma":[0.9871579,0.005548277,0.001676215,0.004209915,0.001161242,0.0002464117],"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":[0.001150629,0.00007631422,0.0001019907,0.0007189208,0.0002979291,0.000001913533,0.004141004,0.0005405003,0.00001642002,0.2337594,0.07503345,0.6841615],"study_design_scores_gemma":[0.003387783,0.0002452446,0.005135177,0.001359098,0.0003220968,0.000002725563,0.1249993,0.0817242,0.0001377202,0.008836158,0.7729941,0.0008563638],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001116996,0.0002726478,0.9239091,0.0401069,0.0008815175,0.002139152,0.003354554,0.000128361,0.02809084],"genre_scores_gemma":[0.1279432,0.001236593,0.7538687,0.02784584,0.000921963,0.0001628552,0.009372514,0.00007342653,0.0785749],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6979606,"threshold_uncertainty_score":0.9998448,"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."}}