{"id":"W2982119098","doi":"10.1016/j.infsof.2019.106205","title":"Automatic prediction of the severity of bugs using stack traces and categorical features","year":2019,"lang":"en","type":"article","venue":"Information and Software Technology","topic":"Software Engineering Research","field":"Computer Science","cited_by":42,"is_retracted":false,"has_abstract":false,"ca_institutions":"Concordia University","funders":"","keywords":"Categorical variable; Computer science; Software bug; Software regression; Data mining; Machine learning; Software; Classifier (UML); Tracing; Predictive modelling; Artificial intelligence; Reliability engineering; Software quality; Software development; Engineering; 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":[0.0007598487,0.001438931,0.0008667785,0.006819852,0.0002824229,0.0008522409,0.001001702,0.001014568,0.001081706],"category_scores_gemma":[0.007131109,0.0004257068,0.001018567,0.002138824,0.0002929209,0.001421452,0.0007721838,0.0008992935,0.0007033036],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003840643,"about_ca_system_score_gemma":0.00117365,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005677462,"about_ca_topic_score_gemma":0.009116802,"domain_scores_codex":[0.999008,0.0001124435,0.0001080174,0.0002616154,0.0003992009,0.0001107756],"domain_scores_gemma":[0.9878696,0.004870513,0.002806653,0.0009710757,0.00268527,0.0007969189],"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":[0.001743112,0.0009099254,0.4665534,0.0007627868,0.0003994863,0.0009156463,0.000264818,0.04916444,0.07072569,0.00166299,0.008232333,0.3986655],"study_design_scores_gemma":[0.00008482918,0.0005645411,0.0899443,0.00005787859,0.0001782326,0.0004733323,0.0001133078,0.8863729,0.01771018,0.003193886,0.001233132,0.00007355819],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8345205,0.0008182953,0.1401029,0.0001874577,0.0001078883,0.0001247478,0.005647275,0.01767951,0.0008114536],"genre_scores_gemma":[0.9373955,0.0001520127,0.05739589,0.00002585186,0.00003712581,0.00003950899,0.004200776,0.0001853879,0.0005679897],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006819852,"threshold_uncertainty_score":0.01128882,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.007654808858986597,"score_gpt":0.2244289516265802,"score_spread":0.2167741427675936,"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."}}