{"id":"W2807522649","doi":"10.1186/s13643-018-0707-8","title":"Technology-assisted title and abstract screening for systematic reviews: a retrospective evaluation of the Abstrackr machine learning tool","year":2018,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":156,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Medical education; Medical physics; Data science","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.2489694,0.0003545485,0.006449623,0.0002884515,0.0002372105,0.0002527802,0.001094421,0.000128678,0.001961489],"category_scores_gemma":[0.2183519,0.0001295027,0.001512462,0.001332163,0.0001185516,0.0001352913,0.0001073293,0.0001779665,0.002217211],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007058633,"about_ca_system_score_gemma":0.00004697339,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002743033,"about_ca_topic_score_gemma":0.00001526183,"domain_scores_codex":[0.9691246,0.0131147,0.01341864,0.0007468672,0.003358765,0.0002364757],"domain_scores_gemma":[0.9764645,0.002675004,0.01521854,0.003273608,0.002311037,0.0000573804],"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":[0.000008283787,0.000106191,0.004542476,0.9027061,0.001075794,7.916034e-7,0.001403239,0.00001076425,0.0007424051,0.006721878,0.04335672,0.03932539],"study_design_scores_gemma":[0.002127893,0.0005255002,0.01391912,0.5605094,0.01613993,0.0003557219,0.003058041,0.1426173,0.0002675158,0.02536425,0.2331405,0.001974804],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.03848819,0.419744,0.09734181,0.001653595,0.003547043,0.3315813,0.0001249876,0.0001437277,0.1073753],"genre_scores_gemma":[0.9616401,0.0005447517,0.0134809,0.0001237799,0.0002357016,0.006339414,0.000008423543,0.00005740705,0.01756957],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9231519,"threshold_uncertainty_score":0.9989508,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6430911952326538,"score_gpt":0.5195480576045787,"score_spread":0.1235431376280751,"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."}}