{"id":"W2994598966","doi":"10.1109/icsme.2019.00018","title":"Improving Bug Triaging with High Confidence Predictions at Ericsson","year":2019,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Computer science; Triage; Software bug; Categorical variable; Context (archaeology); Classifier (UML); Precision and recall; Replicate; Software regression; Machine learning; Artificial intelligence; Data mining; Software; Software quality; Statistics; Software development","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":[0.0002379414,0.00009752511,0.0001067439,0.0001111425,0.00008253773,0.0001442945,0.0005924183,0.0000331209,0.0001128035],"category_scores_gemma":[0.0001403927,0.00007763733,0.00002191623,0.000444902,0.00002155165,0.0004486017,0.0003308143,0.0001507871,0.0002996426],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001291018,"about_ca_system_score_gemma":0.00008717438,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003062077,"about_ca_topic_score_gemma":0.0000129,"domain_scores_codex":[0.9987981,0.00001832655,0.0001110087,0.0003611572,0.0003944745,0.0003169578],"domain_scores_gemma":[0.9986659,0.0004972177,0.00003153882,0.0006235651,0.00008287981,0.00009890238],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001376161,0.0002183494,0.7000446,0.0003765768,0.0002060691,0.0001729347,0.00255704,0.05571409,0.06352866,0.1388908,0.007078575,0.03107464],"study_design_scores_gemma":[0.002342731,0.0007570327,0.1926852,0.0001366307,0.00001257211,0.0001783304,0.00008148493,0.7659591,0.0336437,0.0003681289,0.00304006,0.0007950172],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2843106,0.00002338079,0.7138186,0.000252034,0.000297454,0.0001759586,6.88834e-7,0.000588768,0.0005325127],"genre_scores_gemma":[0.9560493,0.000002505617,0.03930878,0.00005181724,0.00003752435,0.00001884932,8.960116e-7,0.00001150348,0.004518819],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.710245,"threshold_uncertainty_score":0.38514,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.00884423578975883,"score_gpt":0.2223332910563577,"score_spread":0.2134890552665988,"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."}}