{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001470293,0.0001446619,0.0002801839,0.0005017585,0.0004862147,0.00001870748,0.00006540537,0.0002192331,0.00010961],"category_scores_gemma":[0.0001235815,0.0001455916,0.00003936203,0.0006843195,0.00009308704,0.0004768464,0.00005117493,0.0003194599,0.00002205314],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003676271,"about_ca_system_score_gemma":0.0002766267,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002292562,"about_ca_topic_score_gemma":0.00007201882,"domain_scores_codex":[0.9976742,0.000303131,0.001029952,0.0003378617,0.0003118317,0.0003430507],"domain_scores_gemma":[0.9987236,0.0005023152,0.0002256887,0.0001745079,0.000296265,0.00007757383],"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":[0.003455272,0.0002153585,0.0700316,0.01373368,0.000128822,0.000002085292,0.01022371,0.001666918,0.009029761,0.1695266,0.00154757,0.7204387],"study_design_scores_gemma":[0.002133853,0.0001413459,0.03360887,0.0006189392,0.00003262799,0.000005696847,0.002349898,0.9571655,0.001775678,0.0007383388,0.00128047,0.0001487694],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9417786,0.0007380044,0.04920215,0.0004034854,0.001947432,0.00258184,0.0001686918,0.0001757106,0.003004089],"genre_scores_gemma":[0.9904662,0.0003132003,0.008033489,0.00001880753,0.0002272983,0.0001208349,0.0005281855,0.00001961532,0.0002724162],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9554986,"threshold_uncertainty_score":0.5937055,"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."}}