{"id":"W3113997231","doi":"10.1017/s0266462320002159","title":"Creating enriched training sets of eligible studies for large systematic reviews: the utility of PubMed's Best Match algorithm","year":2020,"lang":"en","type":"article","venue":"International Journal of Technology Assessment in Health Care","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"BC Children's Hospital; University of Ottawa; University of Prince Edward Island; University of British Columbia; Children's Hospital of Eastern Ontario","funders":"","keywords":"Ranking (information retrieval); Computer science; sort; Set (abstract data type); False positive paradox; Best practice; Systematic review; Sorting; MEDLINE; Information retrieval; Medicine; Data mining; Algorithm; Machine learning; Biology","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2128217,0.00266742,0.007452608,0.02430145,0.003604112,0.007296282,0.003657198,0.003110418,0.008786614],"category_scores_gemma":[0.5172337,0.003113111,0.008482114,0.01534911,0.002133154,0.006635805,0.006321013,0.002972392,0.002007579],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002795601,"about_ca_system_score_gemma":0.01500661,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002531379,"about_ca_topic_score_gemma":0.007630108,"domain_scores_codex":[0.8126748,0.1427516,0.02705023,0.008551339,0.008105639,0.0008664462],"domain_scores_gemma":[0.3756297,0.5639231,0.01930458,0.02349775,0.01609747,0.00154736],"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":[0.0070185,0.001002642,0.04460242,0.06389168,0.02141307,0.001004184,0.004445474,0.03850127,0.006317641,0.01178187,0.03879036,0.7612309],"study_design_scores_gemma":[0.01316226,0.005919803,0.05261158,0.04583558,0.03861486,0.002980874,0.004302467,0.4691117,0.02318931,0.1973176,0.145164,0.001789915],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08773587,0.0198543,0.8212338,0.00832639,0.001156965,0.0291838,0.01495218,0.01071808,0.006838539],"genre_scores_gemma":[0.1051462,0.001207263,0.8763341,0.000802136,0.0001677324,0.0120027,0.003634447,0.0003608607,0.0003446956],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7871783,"threshold_uncertainty_score":0.970731,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.703124998468772,"score_gpt":0.6088499734728356,"score_spread":0.09427502499593632,"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."}}