{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002686674,0.002109753,0.0009658345,0.002385636,0.0004554765,0.001363127,0.001670191,0.00107446,0.001563655],"category_scores_gemma":[0.01877892,0.0005152438,0.0009416766,0.0007507463,0.0008468122,0.00110881,0.0008386045,0.001186564,0.0004446117],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001284571,"about_ca_system_score_gemma":0.002563704,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007330516,"about_ca_topic_score_gemma":0.01337009,"domain_scores_codex":[0.9966889,0.001140706,0.0002378714,0.0006657178,0.0009524675,0.0003143636],"domain_scores_gemma":[0.9846843,0.01097163,0.001255096,0.0006957081,0.001998793,0.0003945157],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001203996,0.001847631,0.05548085,0.001034543,0.0003171425,0.002655187,0.0008498651,0.2833122,0.1545066,0.004644018,0.006587507,0.4875605],"study_design_scores_gemma":[0.0000906081,0.0004424285,0.008212627,0.00007136894,0.0001030404,0.0005097471,0.0002482874,0.9358744,0.04847107,0.003999079,0.001934357,0.00004302031],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5813532,0.0006160451,0.4044247,0.0005360108,0.00009043863,0.0008616212,0.0004697287,0.008067633,0.003580646],"genre_scores_gemma":[0.8307985,0.0001090374,0.1656104,0.000308471,0.00002143593,0.0003053209,0.001233699,0.0003619053,0.001251289],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007330516,"threshold_uncertainty_score":0.01457566,"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."}}