{"id":"W2953029151","doi":"10.1186/s13643-019-1062-0","title":"A question of trust: can we build an evidence base to gain trust in systematic review automation technologies?","year":2019,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":162,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital","funders":"National Institute of Environmental Health Sciences; Medical Research Council; University College London; U.S. Environmental Protection Agency","keywords":"Automation; Systematic review; Workflow; Documentation; Computer science; Disk formatting; Data science; Medicine; Knowledge management; Risk analysis (engineering); Software engineering; MEDLINE; Engineering","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":[0.8650245,0.003866956,0.01785391,0.04168212,0.006719077,0.03941325,0.01574671,0.02642107,0.01007662],"category_scores_gemma":[0.970049,0.009091217,0.01232397,0.02688787,0.04093296,0.07398723,0.02406472,0.02848829,0.004424243],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03946395,"about_ca_system_score_gemma":0.08839786,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009874507,"about_ca_topic_score_gemma":0.007468136,"domain_scores_codex":[0.07409177,0.6912243,0.1624352,0.01718425,0.05192677,0.003137724],"domain_scores_gemma":[0.007796614,0.8482038,0.0527806,0.03866113,0.04986132,0.002696517],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002980126,0.0002656027,0.01070476,0.3511498,0.01983118,0.0006387616,0.02998558,0.004035554,0.0007929821,0.1474061,0.1177766,0.3144329],"study_design_scores_gemma":[0.002251713,0.0007692885,0.004454084,0.5200628,0.007093895,0.0007503505,0.004925456,0.009490979,0.000909857,0.3363936,0.1119902,0.0009077128],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.004984786,0.1563014,0.1184885,0.682478,0.02143416,0.006791127,0.001705328,0.001087899,0.006728816],"genre_scores_gemma":[0.2586682,0.07792933,0.3704409,0.230861,0.01777879,0.03954304,0.001985716,0.001099419,0.00169369],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1349755,"threshold_uncertainty_score":0.2863324,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4757505986788351,"score_gpt":0.4983922702418923,"score_spread":0.02264167156305724,"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."}}