{"id":"W4401386467","doi":"10.1080/03155986.2024.2381306","title":"Optimization modeling and verification from problem specifications using a multi-agent multi-stage LLM framework","year":2024,"lang":"en","type":"article","venue":"INFOR Information Systems and Operational Research","topic":"Business Process Modeling and Analysis","field":"Business, Management and Accounting","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; Huawei Technologies (Canada)","funders":"","keywords":"Computer science; Task (project management); Modeling language; Solver; Representation (politics); Artificial intelligence; Optimization problem; Machine learning; Programming language; Systems engineering; Algorithm","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.006768789,0.001246962,0.0007243697,0.001495138,0.0009976166,0.003060024,0.002762725,0.001310174,0.005819446],"category_scores_gemma":[0.017882,0.0009443419,0.002621568,0.0008174112,0.001439179,0.003446713,0.003644593,0.002988977,0.001001765],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002530641,"about_ca_system_score_gemma":0.005543712,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009814909,"about_ca_topic_score_gemma":0.01980012,"domain_scores_codex":[0.9951211,0.002429563,0.000431449,0.000551444,0.001248261,0.0002181672],"domain_scores_gemma":[0.9900485,0.00644179,0.0008505616,0.001587503,0.0008526896,0.0002188681],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001370912,0.0002393816,0.001465206,0.0004827081,0.0001201416,0.0003276707,0.0005725542,0.7114465,0.004844925,0.1802427,0.00413814,0.09598296],"study_design_scores_gemma":[0.00002145996,0.00002423203,0.00007011765,0.0000315127,0.00001474699,0.00002824872,0.00004269341,0.9624193,0.001956268,0.03079992,0.004580406,0.00001110443],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002628627,0.00005571138,0.9938508,0.0003091901,0.00001372802,0.0001339225,0.0001448448,0.001656468,0.001206552],"genre_scores_gemma":[0.0768102,0.00008955637,0.9206479,0.0001676603,0.00001637285,0.0003626023,0.0005173715,0.0002984338,0.00108981],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.009814909,"threshold_uncertainty_score":0.03579724,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2535615152377986,"score_gpt":0.3732107720741901,"score_spread":0.1196492568363914,"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."}}