{"id":"W4403536522","doi":"10.1145/3691620.3695061","title":"What Makes a High-Quality Training Dataset for Large Language Models: A Practitioners' Perspective","year":2024,"lang":"en","type":"article","venue":"","topic":"Data Quality and Management","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"Natural Science Foundation of Chongqing","keywords":"Perspective (graphical); Computer science; Training (meteorology); Quality (philosophy); Language model; Artificial intelligence; Natural language processing; Data science; Machine learning","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":["scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007759336,0.000141209,0.000247765,0.0002292775,0.0001733789,0.002450913,0.0005481323,0.00005396311,0.001251225],"category_scores_gemma":[0.001731733,0.0001014434,0.000118117,0.0004377294,0.00004646418,0.005118967,0.0002529614,0.0001200526,0.0004504488],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007474463,"about_ca_system_score_gemma":0.0000813992,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004343489,"about_ca_topic_score_gemma":0.001056964,"domain_scores_codex":[0.9972723,0.0002625292,0.0004745153,0.0007760278,0.0009011271,0.0003135724],"domain_scores_gemma":[0.9968282,0.002097692,0.00009367001,0.0007434677,0.0001449034,0.00009206033],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002873563,0.0000630061,1.794416e-7,0.00002169471,0.00004881874,0.00001377763,0.01447453,0.00004149413,0.00003300914,0.7428566,0.2332117,0.009206475],"study_design_scores_gemma":[0.0002452862,0.00003354603,0.000008453453,0.00002747318,0.00002367945,0.000002358922,0.3527008,0.008376753,0.00004163216,0.2838537,0.3545495,0.0001367709],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.003196368,0.001450024,0.8999192,0.03748109,0.001930242,0.001006431,0.04087043,0.0003122441,0.01383402],"genre_scores_gemma":[0.9351417,0.0001819806,0.01875039,0.009159809,0.0004716842,0.000278769,0.008988736,0.00003573791,0.02699113],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9319454,"threshold_uncertainty_score":0.9996617,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3320277085142438,"score_gpt":0.5121126675199051,"score_spread":0.1800849590056613,"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."}}