{"id":"W3136275640","doi":"10.1109/bigdata50022.2020.9378201","title":"Customizing Contextualized Language Models for Legal Document Reviews","year":2020,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":25,"is_retracted":false,"has_abstract":true,"ca_institutions":"Thomson Reuters (Canada)","funders":"","keywords":"Computer science; Natural language processing; Domain (mathematical analysis); Artificial intelligence; Inference; Focus (optics); Task (project management); Security token; Language model; Question answering","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":[],"consensus_categories":[],"category_scores_codex":[0.0002889472,0.00009991259,0.0002210254,0.0000194318,0.000049457,0.0001349454,0.0005436058,0.00002687941,0.00003033512],"category_scores_gemma":[0.0000650998,0.00007963305,0.0001005393,0.00008825921,0.000006577653,0.0005596788,0.0001684793,0.00005941608,0.0000504127],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002082579,"about_ca_system_score_gemma":0.00002813404,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005066846,"about_ca_topic_score_gemma":0.00000692284,"domain_scores_codex":[0.9990366,0.00004156082,0.0002770889,0.0003258824,0.0001223729,0.0001965091],"domain_scores_gemma":[0.9994418,0.00004817774,0.0000595801,0.0003153363,0.00002838185,0.000106728],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001134273,0.000009625123,0.000001234331,0.00006144366,0.00001075134,0.000005369359,0.004362982,0.001957213,0.001421138,0.8719929,0.008954039,0.1112119],"study_design_scores_gemma":[0.000604765,0.00002710455,1.497989e-7,0.00001532602,0.000004284786,0.000001711507,0.00009510224,0.9010053,0.001004547,0.002288751,0.09483071,0.0001222181],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0003784479,0.001145451,0.9854516,0.005468709,0.0001107602,0.0004922213,7.40653e-7,0.0001920723,0.006759996],"genre_scores_gemma":[0.3146446,0.00005774215,0.6704412,0.01269143,0.0001921995,0.00009788434,0.000002074142,0.00001126816,0.001861585],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8990481,"threshold_uncertainty_score":0.3247342,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0689026083376294,"score_gpt":0.3120019498877744,"score_spread":0.243099341550145,"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."}}