{"id":"W3091435513","doi":"10.1145/3365438.3410953","title":"Leveraging natural-language requirements for deriving better acceptance criteria from models","year":2020,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"Natural Sciences and Engineering Research Council of Canada; Fonds National de la Recherche Luxembourg","keywords":"Computer science; Requirements analysis; Natural language; Software requirements specification; Non-functional testing; Requirements elicitation; Acceptance testing; Software engineering; Information system; System requirements specification; Software requirements; System requirements; Software; Non-functional requirement; Information retrieval; Software system; Software development; Artificial intelligence; Programming language; Software design; Engineering; Software construction","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0001214085,0.0001382535,0.0001434452,0.00004642558,0.00007444812,0.0003074274,0.001067571,0.00003588593,0.00008266047],"category_scores_gemma":[0.000286611,0.0001327206,0.00005830214,0.0002155161,0.0000115719,0.001043947,0.0004401426,0.0001446313,0.00002816059],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004769925,"about_ca_system_score_gemma":0.00002375728,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002589061,"about_ca_topic_score_gemma":0.000002285431,"domain_scores_codex":[0.9986248,0.0000209423,0.0001694938,0.0004938534,0.0003140224,0.0003768792],"domain_scores_gemma":[0.9990182,0.0003732789,0.0000271444,0.0003997873,0.00005846681,0.0001231433],"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":[0.0001857528,0.0001074611,0.009820255,0.0004588063,0.0004084695,0.0002535833,0.05792973,0.007625571,0.3955252,0.003590925,0.06686555,0.4572287],"study_design_scores_gemma":[0.0003296031,0.00004210535,0.002012884,0.00002493251,0.000002425686,8.464332e-7,0.00004537821,0.9796484,0.0165115,0.0008346996,0.00033159,0.0002156386],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1657212,0.0002565589,0.8304839,0.002505257,0.0003073671,0.0001803868,0.000003509675,0.0004625496,0.00007932018],"genre_scores_gemma":[0.7201864,0.000001520938,0.277389,0.002154821,0.0001564988,0.00002545842,0.000004101426,0.00001562916,0.00006659789],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9720228,"threshold_uncertainty_score":0.5412191,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06341128614539714,"score_gpt":0.316385960359451,"score_spread":0.2529746742140538,"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."}}