{"id":"W2973296283","doi":"10.1109/tse.2019.2941880","title":"Studying the Impact of Noises in Build Breakage Data","year":2019,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Computer science; Replicate; Breakage; Server; Troubleshooting; Timeout; Data mining; Data science; World Wide Web; Statistics; Operating 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":[0.02222134,0.001711993,0.001188432,0.005724118,0.001174456,0.003029471,0.002231496,0.00254141,0.0007642843],"category_scores_gemma":[0.1202193,0.001069687,0.002215594,0.007275775,0.002283273,0.004868251,0.002670206,0.003545919,0.0006528511],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001913406,"about_ca_system_score_gemma":0.001831199,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01886236,"about_ca_topic_score_gemma":0.02074496,"domain_scores_codex":[0.9741772,0.00901694,0.002771368,0.006825145,0.006055758,0.001153546],"domain_scores_gemma":[0.7896253,0.1612335,0.01771839,0.02153644,0.00810322,0.001783013],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006300358,0.0004946896,0.8586049,0.0006574276,0.0007156291,0.0007393794,0.002441871,0.08708007,0.002372897,0.002762937,0.00686986,0.03663018],"study_design_scores_gemma":[0.00011995,0.000503389,0.6093639,0.0004754022,0.0005280126,0.001469314,0.003459386,0.3390227,0.007897339,0.0105459,0.02637993,0.0002347391],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9264106,0.002296367,0.04605517,0.001454841,0.0002719241,0.0002468513,0.01977948,0.001492867,0.001991988],"genre_scores_gemma":[0.9183797,0.0005034966,0.0273627,0.0004347799,0.0001542262,0.0002174577,0.05196543,0.0002687647,0.0007133158],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02222134,"threshold_uncertainty_score":0.1175191,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02816949054392912,"score_gpt":0.2874072299796261,"score_spread":0.2592377394356969,"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."}}