{"id":"W4413362641","doi":"10.3389/fninf.2025.1609077","title":"Large language models can extract metadata for annotation of human neuroimaging publications","year":2025,"lang":"en","type":"article","venue":"Frontiers in Neuroinformatics","topic":"Topic Modeling","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"National Institute on Drug Abuse","keywords":"Annotation; Metadata; Computer science; Information retrieval; USable; Natural language processing; Neuroimaging; Artificial intelligence; Data science; World Wide Web; Psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.0003323679,0.0001016602,0.0001863723,0.0004719954,0.0000948043,0.0001571647,0.000822742,0.00003373745,4.051344e-7],"category_scores_gemma":[0.0001380229,0.000110159,0.00005204956,0.0005079344,0.00001986745,0.00202303,0.0001760375,0.0001298837,1.844328e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003567387,"about_ca_system_score_gemma":0.00009229279,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001415606,"about_ca_topic_score_gemma":0.00001613514,"domain_scores_codex":[0.9988803,0.00002833045,0.0005218496,0.0001715789,0.0001607697,0.0002371376],"domain_scores_gemma":[0.998924,0.00006123543,0.0001791514,0.0007176688,0.00008433344,0.00003357808],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000009603671,0.0002369515,0.004157681,0.0006395351,0.0000552979,0.000003784558,0.01833745,0.04914758,0.0007523465,0.8243135,0.02220319,0.08014314],"study_design_scores_gemma":[0.0003633402,0.00001263103,0.0003488863,0.0000224139,0.000008857649,8.004229e-7,0.0005330684,0.9862341,0.000323699,0.01066709,0.001402589,0.00008254198],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01276892,0.00008586689,0.9837015,0.0007421155,0.0004028542,0.0003480135,0.00003613315,0.00007632424,0.001838255],"genre_scores_gemma":[0.5618446,0.00001546793,0.4370079,0.0006059698,0.00001309139,0.00003601719,0.00005564477,0.000008978688,0.0004122987],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9370865,"threshold_uncertainty_score":0.4492152,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02792275338671165,"score_gpt":0.2913209766535713,"score_spread":0.2633982232668596,"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."}}