{"id":"W4389161815","doi":"10.1145/3611643.3613872","title":"Prioritizing Natural Language Test Cases Based on Highly-Used Game Features","year":2023,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Computer science; Test case; Test (biology); Code coverage; Machine learning; Test suite; Test Management Approach; Natural language; System under test; Artificial intelligence; Data mining; Software; Programming language; Software system","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":[0.0002548248,0.0001623053,0.0001423339,0.0002580416,0.0001142045,0.0002201111,0.0005919196,0.00006009005,0.000004137234],"category_scores_gemma":[0.001606216,0.0001308393,0.00006577378,0.0007892587,0.00002932486,0.0001455582,0.0001527219,0.0002019087,0.0001113543],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003804233,"about_ca_system_score_gemma":0.00004909017,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002002539,"about_ca_topic_score_gemma":0.00001925926,"domain_scores_codex":[0.9987865,0.00004013421,0.0001308049,0.0003849849,0.0003210531,0.000336541],"domain_scores_gemma":[0.9966127,0.002597848,0.00004301193,0.0006292625,0.0000453442,0.00007186088],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000283435,0.000374335,0.02542683,0.0001546871,0.00002770136,0.006676792,0.003260696,0.0002021724,0.01611404,0.01066446,0.7351269,0.2019431],"study_design_scores_gemma":[0.001599241,0.001300849,0.1313211,0.0007911511,0.00002249884,0.000487602,0.0001061592,0.7960998,0.05328202,0.00862221,0.004362362,0.002004932],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"software","genre_gemma":"empirical","genre_scores_codex":[0.4276145,0.0008784602,0.08936352,0.01600413,0.003039013,0.001224912,0.00003789673,0.4475589,0.01427872],"genre_scores_gemma":[0.9363222,0.000001325969,0.06103351,0.001305973,0.00008725492,0.0000159594,0.000008677414,0.00001601247,0.001209085],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7958977,"threshold_uncertainty_score":0.5335471,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01845999358637851,"score_gpt":0.2859660803773075,"score_spread":0.267506086790929,"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."}}