{"id":"W2109294629","doi":"10.1007/s00766-004-0204-6","title":"A framework for empirical evaluation of conceptual modeling techniques","year":2004,"lang":"en","type":"article","venue":"Requirements Engineering","topic":"Business Process Modeling and Analysis","field":"Business, Management and Accounting","cited_by":228,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia; Simon Fraser University","funders":"","keywords":"Computer science; Categorical variable; Categorization; Grammar; Rule-based machine translation; Empirical research; Task (project management); Modeling language; Variables; Machine learning; Artificial intelligence; Natural language processing; Systems engineering; Engineering; Mathematics; Programming language; Software; Statistics; Linguistics","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":[0.0006567477,0.0001502369,0.000216702,0.0002451898,0.00006750286,0.00005373007,0.0001529905,0.00008607437,0.00001636684],"category_scores_gemma":[0.0003886684,0.0001489735,0.0001029737,0.0003912487,0.00002213832,0.0005651437,0.00004962058,0.00008090492,0.000004005195],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007235845,"about_ca_system_score_gemma":0.00003964472,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005410667,"about_ca_topic_score_gemma":0.000001721202,"domain_scores_codex":[0.9987601,0.000002375247,0.0003374068,0.0002238019,0.0004578847,0.0002184232],"domain_scores_gemma":[0.9991449,0.00001587923,0.0001243799,0.0001645821,0.0005404069,0.000009841121],"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":[0.0000111825,0.00005749026,0.000170943,0.0002164859,0.00005579719,2.296619e-7,0.00008583543,0.9636974,0.001919577,0.02855938,0.000008345061,0.005217292],"study_design_scores_gemma":[0.0004646964,0.000006560072,0.00001589999,0.0003161034,0.0002193825,2.202288e-7,0.0001155584,0.9727449,0.001191823,0.02453524,0.0002093918,0.0001802713],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2428972,0.0001404921,0.7561898,0.0001532122,0.0001017902,0.0002145339,0.000001132671,0.0001496133,0.0001522055],"genre_scores_gemma":[0.962221,0.000005292156,0.03697448,0.0001305631,0.0005197379,0.00009464109,0.00002065885,0.00003153879,0.000002053991],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7193239,"threshold_uncertainty_score":0.6074965,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1449234842993098,"score_gpt":0.3523727618234092,"score_spread":0.2074492775240994,"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."}}