{"id":"W4379469178","doi":"10.1186/s13643-023-02247-9","title":"Ensemble of deep learning language models to support the creation of living systematic reviews for the COVID-19 literature","year":2023,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institutes of Health Research; Horizon 2020 Framework Programme; European Commission; Innosuisse - Schweizerische Agentur für Innovationsförderung; Université de Genève; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Science Foundation","keywords":"Artificial intelligence; Machine learning; Ensemble forecasting; Computer science; Ranking (information retrieval); Task (project management); Natural language processing; Ensemble learning; Class (philosophy); Triage; Medicine; Data science","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02212878,0.001860763,0.001964633,0.006003147,0.0006522497,0.002535986,0.002238719,0.001632507,0.001534024],"category_scores_gemma":[0.05792317,0.0007579041,0.003625018,0.002608275,0.0003778763,0.002708186,0.002008462,0.001972413,0.001281421],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001961627,"about_ca_system_score_gemma":0.005971439,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007998585,"about_ca_topic_score_gemma":0.02798745,"domain_scores_codex":[0.9908331,0.005502192,0.001238556,0.001329684,0.0009078435,0.0001886228],"domain_scores_gemma":[0.9608485,0.02858008,0.003126476,0.002163956,0.004654797,0.000626265],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001848019,0.0006756323,0.0448246,0.0104842,0.008746147,0.000903285,0.0009854689,0.1643697,0.006157185,0.002857144,0.04041496,0.7177335],"study_design_scores_gemma":[0.000379552,0.0006335452,0.006177025,0.002491492,0.003264656,0.0003572846,0.0002189397,0.9566923,0.00377954,0.01230616,0.0135814,0.0001181234],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2270151,0.08021929,0.6195482,0.0134152,0.002098363,0.003467366,0.02518374,0.02240844,0.006644243],"genre_scores_gemma":[0.5267293,0.007331129,0.4351444,0.003110455,0.000602532,0.002374404,0.02148321,0.0004365899,0.002788137],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9778712,"threshold_uncertainty_score":0.1170295,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6019366327301586,"score_gpt":0.5165044718164059,"score_spread":0.08543216091375272,"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."}}