{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009457923,0.0001464348,0.0001814572,0.000394145,0.0003199775,0.0005007908,0.0007458408,0.0001205706,0.000001991335],"category_scores_gemma":[0.0002891851,0.0001458087,0.00009769045,0.0005747566,0.00001526585,0.004577052,0.00001538087,0.0001755424,0.0002048788],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007703271,"about_ca_system_score_gemma":0.0001588477,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001152514,"about_ca_topic_score_gemma":0.000004654075,"domain_scores_codex":[0.9983472,0.00005018944,0.0005984201,0.0002165099,0.000482953,0.0003047128],"domain_scores_gemma":[0.9980779,0.0005123182,0.0003919024,0.0006592398,0.0002659616,0.00009263639],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00005121307,0.00002629637,0.000001428421,0.0001885879,0.0000500794,0.000002040229,0.03348321,0.7803218,0.00007470679,0.006449765,0.0000969495,0.179254],"study_design_scores_gemma":[0.0005398752,0.00007984845,0.000002010345,0.00005390526,0.00001260199,0.00002793047,0.006578961,0.9803803,0.0004713607,0.0005670862,0.01111852,0.0001676012],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.007504959,0.00001593477,0.9879549,0.000755369,0.001669988,0.000628518,0.00003830377,0.0006039485,0.0008281048],"genre_scores_gemma":[0.9672398,0.00002245314,0.03183729,0.0001426118,0.00009359882,0.0001452319,0.00002035962,0.00001097502,0.0004876855],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9597349,"threshold_uncertainty_score":0.5945907,"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."}}