{"id":"W4404280355","doi":"10.1016/j.autcon.2024.105846","title":"Performance comparison of retrieval-augmented generation and fine-tuned large language models for construction safety management knowledge retrieval","year":2024,"lang":"en","type":"article","venue":"Automation in Construction","topic":"Occupational Health and Safety Research","field":"Health Professions","cited_by":90,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Computer science; Information retrieval","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":[],"consensus_categories":[],"category_scores_codex":[0.00227634,0.001930491,0.001810925,0.001707361,0.0005886494,0.002262286,0.00217389,0.002266511,0.008497347],"category_scores_gemma":[0.009033238,0.000558473,0.001344065,0.00124783,0.0003727494,0.003073483,0.001730927,0.001332976,0.004814279],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009624968,"about_ca_system_score_gemma":0.002021636,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01924532,"about_ca_topic_score_gemma":0.01291668,"domain_scores_codex":[0.9980123,0.0007011062,0.0002286379,0.0005544171,0.0002975374,0.0002059488],"domain_scores_gemma":[0.9951835,0.003304197,0.0001171209,0.0005358645,0.0006432226,0.0002160144],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006438755,0.00125243,0.004276533,0.00143383,0.0006981661,0.0005162758,0.0004023272,0.06602301,0.03783762,0.001197714,0.02519585,0.8547274],"study_design_scores_gemma":[0.0006095567,0.001302322,0.00396516,0.0000646241,0.0006208667,0.0005257233,0.0004469963,0.9509,0.03356856,0.001620219,0.006240163,0.0001357759],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6848989,0.01086099,0.1791825,0.001253495,0.001158816,0.000682498,0.007342781,0.1011017,0.01351837],"genre_scores_gemma":[0.8215591,0.001546541,0.1458533,0.0005178378,0.0001637593,0.0002547253,0.02104239,0.001447319,0.007615124],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01924532,"threshold_uncertainty_score":0.0382666,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06852945978436745,"score_gpt":0.448781636626997,"score_spread":0.3802521768426296,"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."}}