{"id":"W4385780698","doi":"10.1145/3613447","title":"Toward Best Practices for Training Multilingual Dense Retrieval Models","year":2023,"lang":"en","type":"article","venue":"ACM Transactions on Information Systems","topic":"Topic Modeling","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canada First Research Excellence Fund; Compute Canada","keywords":"Computer science; Transformer; Language model; Architecture; Encoder; Relevance (law); Natural language processing; Artificial intelligence; Transfer of learning; Training set; Information retrieval; Variety (cybernetics); Data science","routes":{"ca_aff":true,"ca_fund":true,"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":[0.01124741,0.001993206,0.001535992,0.002117187,0.0009744164,0.003352187,0.005440544,0.002423336,0.003930444],"category_scores_gemma":[0.031637,0.001845071,0.00123466,0.001799566,0.001361848,0.006750896,0.004347054,0.005186656,0.003966081],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001967122,"about_ca_system_score_gemma":0.003125403,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01061341,"about_ca_topic_score_gemma":0.02508262,"domain_scores_codex":[0.9943998,0.00331949,0.0003953014,0.0008355748,0.0007906802,0.0002590369],"domain_scores_gemma":[0.9873017,0.007896454,0.0003312194,0.002229684,0.001938399,0.0003024852],"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":[0.0004234581,0.0004808957,0.002829313,0.000718314,0.000368907,0.0002229858,0.0007023633,0.2639135,0.007959372,0.03823209,0.01615703,0.6679917],"study_design_scores_gemma":[0.000102784,0.0001040728,0.0002284551,0.0001306106,0.00006537911,0.00009816083,0.0001845796,0.9368683,0.004749035,0.05116259,0.006273232,0.00003282306],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008722712,0.001771996,0.9820004,0.001096535,0.00006364271,0.0001557008,0.0002297942,0.003985376,0.001973799],"genre_scores_gemma":[0.1191272,0.001279623,0.8730247,0.0008610594,0.0001084128,0.0005619187,0.001727281,0.0009376808,0.002372096],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01124741,"threshold_uncertainty_score":0.05948269,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2134837990183883,"score_gpt":0.3481269699009422,"score_spread":0.1346431708825539,"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."}}