{"id":"W3121686920","doi":"10.1186/s13643-021-01583-y","title":"Considerations for conducting systematic reviews: evaluating the performance of different methods for de-duplicating references","year":2021,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":238,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kingston Health Sciences Centre; Queen's University","funders":"","keywords":"Systematic review; Medicine; PsycINFO; Flagging; Information retrieval; Computer science; MEDLINE; Set (abstract data type); Data mining; Database; Data science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"simulation_or_modeling","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5266486,0.0007013317,0.01662136,0.0002079684,0.0008125379,0.00101858,0.001872411,0.0001350247,0.0007116284],"category_scores_gemma":[0.6942765,0.0002527747,0.00478931,0.001224367,0.00009527647,0.0002782633,0.0001761226,0.0002137697,0.0002099478],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001050732,"about_ca_system_score_gemma":0.0002631167,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002598829,"about_ca_topic_score_gemma":0.00001340998,"domain_scores_codex":[0.7805506,0.1604216,0.05267592,0.001919249,0.003631002,0.0008016399],"domain_scores_gemma":[0.5736367,0.3332556,0.06947115,0.01394678,0.009344998,0.0003448229],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000003169206,0.00005859738,0.0002129744,0.9722319,0.0005284967,2.036422e-7,0.00271189,0.00005571224,0.004143563,0.009783096,0.004100061,0.006170392],"study_design_scores_gemma":[0.0009827862,0.0006149113,0.00006229972,0.3107138,0.01270398,0.0004092476,0.01503353,0.6105109,0.004188274,0.03081042,0.01258894,0.001380934],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03138985,0.1722413,0.672404,0.001322616,0.001321559,0.1200418,0.00003910824,0.00002679934,0.001212986],"genre_scores_gemma":[0.1532101,0.002198541,0.7689855,0.0007368524,0.0003106561,0.06713679,0.00002296054,0.00007570746,0.007322868],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.661518,"threshold_uncertainty_score":0.9999924,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9330778224572595,"score_gpt":0.6546697466149455,"score_spread":0.278408075842314,"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."}}