{"id":"W3112337105","doi":"10.1186/s13643-020-01520-5","title":"Aligning text mining and machine learning algorithms with best practices for study selection in systematic literature reviews","year":2020,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; Precision Nanosystems (Canada)","funders":"","keywords":"Medicine; Selection (genetic algorithm); Systematic review; Machine learning; Artificial intelligence; Algorithm; MEDLINE; Data science; Computer 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":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2488507,0.0009819433,0.01744238,0.0005865004,0.0003672331,0.002770536,0.001388935,0.0001519619,0.0001982814],"category_scores_gemma":[0.2610602,0.0003981522,0.001374042,0.004319966,0.00003595264,0.0009607135,0.0001462677,0.0004362526,0.0006757469],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007477558,"about_ca_system_score_gemma":0.00005584228,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000240104,"about_ca_topic_score_gemma":0.0002156985,"domain_scores_codex":[0.8938214,0.07264476,0.02671124,0.002326315,0.00392022,0.000576087],"domain_scores_gemma":[0.9463793,0.01315288,0.03701514,0.002014561,0.001081432,0.0003567028],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0000240618,0.0001795586,0.0143515,0.9541047,0.0005157258,0.00002304588,0.02411264,0.00005802646,0.00005788988,0.0000709325,0.002781827,0.003720121],"study_design_scores_gemma":[0.003012193,0.003171644,0.0002743719,0.5691317,0.01062345,0.0007786386,0.05652903,0.2447219,0.0000112534,0.000103408,0.1093545,0.002287895],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.02017066,0.668683,0.1015931,0.001666232,0.0004726936,0.2049015,0.00002322414,0.00008161621,0.002407993],"genre_scores_gemma":[0.5960148,0.02343168,0.2514726,0.003129048,0.001700205,0.07786605,0.0001034719,0.0005747475,0.04570739],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.6452513,"threshold_uncertainty_score":0.9998471,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6685598516985551,"score_gpt":0.5129697103957964,"score_spread":0.1555901413027587,"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."}}