{"id":"W4400983681","doi":"10.1002/jrsm.1736","title":"Considerations for conducting systematic reviews: A follow‐up study to evaluate the performance of various automated methods for reference de‐duplication","year":2024,"lang":"en","type":"article","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University; Queen's University","funders":"","keywords":"Computer science; Evaluation methods; Research methodology; Reliability engineering; Engineering; Medicine","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":["metaresearch","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8064832,0.0003472852,0.004553088,0.0009969515,0.0008388587,0.001568434,0.002015342,0.0001002984,0.0005908526],"category_scores_gemma":[0.7719032,0.0001504631,0.00134974,0.003416983,0.0001174131,0.0002722951,0.0001962648,0.0003056478,0.0002369354],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001829191,"about_ca_system_score_gemma":0.0005610086,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003212072,"about_ca_topic_score_gemma":0.00002157282,"domain_scores_codex":[0.638258,0.3368992,0.01574238,0.002345132,0.005685149,0.001070151],"domain_scores_gemma":[0.3180937,0.6638458,0.00251845,0.006758223,0.008464498,0.0003193774],"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":[0.0001524125,0.0006570884,0.0002731894,0.07756174,0.004819855,0.000002493126,0.04026252,0.001121953,0.1393987,0.02518779,0.06321147,0.6473507],"study_design_scores_gemma":[0.0001986981,0.000439823,0.0001610471,0.003890978,0.001968582,0.00002258609,0.007199958,0.9395254,0.01838154,0.01459968,0.01330467,0.0003070083],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03120945,0.005886632,0.9337484,0.001619008,0.0003494431,0.02670989,0.00003148782,0.00005670126,0.0003889878],"genre_scores_gemma":[0.2384695,0.00008138674,0.7420619,0.00004725482,0.00005568254,0.0173692,0.000001490321,0.00003192248,0.001881744],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9384035,"threshold_uncertainty_score":0.999468,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9579403401628281,"score_gpt":0.7454122149889704,"score_spread":0.2125281251738578,"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."}}