{"id":"W6894038808","doi":"10.5281/zenodo.8111951","title":"Automated Domain Modeling with Large Language Models: A Comparative Study","year":2023,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Domain (mathematical analysis); Domain model; Automation; Key (lock); Domain analysis","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":[0.008227366,0.000737645,0.0008618597,0.002341557,0.0008109281,0.002664902,0.002015991,0.001052737,0.003415139],"category_scores_gemma":[0.025436,0.0006073311,0.001373245,0.001812419,0.0006793626,0.004778294,0.00214157,0.001527201,0.001204624],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001680699,"about_ca_system_score_gemma":0.001744645,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007779112,"about_ca_topic_score_gemma":0.007951584,"domain_scores_codex":[0.994076,0.003688914,0.0003659826,0.0004877759,0.001247577,0.0001337989],"domain_scores_gemma":[0.9536918,0.03803883,0.0009179162,0.005382308,0.001684377,0.0002847797],"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":[0.001865726,0.001358252,0.01431171,0.002196873,0.0006047014,0.0005789393,0.00314653,0.2198837,0.01309031,0.03628685,0.008930957,0.6977454],"study_design_scores_gemma":[0.0001894048,0.0003392791,0.004358734,0.0001890769,0.0002955125,0.000389935,0.001014761,0.9296418,0.01551679,0.01817601,0.02981668,0.00007204035],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2542402,0.003146447,0.6991324,0.001244362,0.00009703325,0.0007580159,0.002482965,0.02254752,0.01635103],"genre_scores_gemma":[0.6655793,0.001442732,0.3232964,0.0001636152,0.00003569492,0.0002609902,0.005540116,0.001925295,0.001755864],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.008227366,"threshold_uncertainty_score":0.04351097,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07254845640432699,"score_gpt":0.2928745200930867,"score_spread":0.2203260636887597,"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."}}