{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002646115,0.001213178,0.0009151546,0.001538199,0.0004295091,0.001484482,0.001429257,0.001077421,0.002556996],"category_scores_gemma":[0.009445618,0.0006543346,0.001256693,0.0009547104,0.0004031588,0.002591275,0.001221038,0.001526131,0.003246621],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007760955,"about_ca_system_score_gemma":0.001787652,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006303597,"about_ca_topic_score_gemma":0.01314614,"domain_scores_codex":[0.998144,0.0008084584,0.0001361117,0.0006078283,0.000177075,0.0001264728],"domain_scores_gemma":[0.9963504,0.001819087,0.000244251,0.0006098951,0.0008234115,0.000152921],"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":[0.0009252877,0.0006299975,0.007564744,0.001018134,0.0006920273,0.000502992,0.001040429,0.2583417,0.05919406,0.006898697,0.03342052,0.6297714],"study_design_scores_gemma":[0.00004625674,0.0001082896,0.001074707,0.0000334497,0.0001063269,0.0001304188,0.0001477708,0.9721594,0.01349143,0.003729318,0.008922392,0.00005018364],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1249747,0.002637936,0.8303411,0.0009384929,0.0004900122,0.0005100197,0.00229738,0.03348135,0.004328936],"genre_scores_gemma":[0.5708719,0.001070055,0.4127473,0.0004997234,0.0003840702,0.0007065887,0.007989022,0.001881044,0.003850211],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.006303597,"threshold_uncertainty_score":0.01399416,"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."}}