{"id":"W4407986251","doi":"10.1007/s10664-025-10619-z","title":"An empirical study on LLM-based classification of requirements-related provisions in food-safety regulations","year":2025,"lang":"en","type":"article","venue":"Empirical Software Engineering","topic":"Technology and Data Analysis","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Ottawa","funders":"Natural Sciences and Engineering Research Council of Canada; Ontario Centre of Innovation","keywords":"Food safety; Empirical research; Business; Computer science; Food science; Chemistry; Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.01193581,0.0003491036,0.0003167904,0.00290462,0.0009358088,0.001552583,0.001420875,0.0008530953,0.004554838],"category_scores_gemma":[0.06911366,0.0002097275,0.0007343247,0.003103297,0.001183639,0.002660087,0.001635696,0.001905188,0.000761725],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001954341,"about_ca_system_score_gemma":0.001577077,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01096864,"about_ca_topic_score_gemma":0.01115756,"domain_scores_codex":[0.989749,0.006477867,0.0007656871,0.0008712402,0.001510921,0.0006252006],"domain_scores_gemma":[0.8430997,0.1209858,0.01951589,0.00479085,0.009512971,0.002094731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003008273,0.001319373,0.9791504,0.00005861076,0.00005232443,0.00008947332,0.002293322,0.0005175954,0.0004635535,0.001030899,0.0003784316,0.01434514],"study_design_scores_gemma":[0.00003717393,0.0005033804,0.9713657,0.00009007306,0.00009330855,0.0001375521,0.0132624,0.01138499,0.0009049457,0.0005697025,0.001627129,0.00002342724],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9981284,0.00003946087,0.0005404312,0.00006040124,0.00000388087,0.00002929257,0.0000969792,0.000004740353,0.001096503],"genre_scores_gemma":[0.9984786,0.00002193157,0.0007057786,0.000030511,0.000005061603,0.00003416653,0.0003337463,0.000005041177,0.0003851208],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01193581,"threshold_uncertainty_score":0.06312329,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03242033749613746,"score_gpt":0.3338891427159063,"score_spread":0.3014688052197688,"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."}}