{"id":"W4401071165","doi":"10.1109/icosse62619.2024.00010","title":"Identifying and Detecting Inaccurate Stack Traces in Bug Reports","year":2024,"lang":"en","type":"article","venue":"","topic":"Advanced Malware Detection Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Lethbridge","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Stack (abstract data type); Computer science; Programming language","routes":{"ca_aff":true,"ca_fund":true,"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":[0.0003838516,0.00008826076,0.00009171978,0.0002394102,0.00005530498,0.0003852246,0.0001254566,0.00003870345,0.000007006839],"category_scores_gemma":[0.00007137411,0.00008159726,0.00002003688,0.0005295106,0.00001748338,0.001345386,0.0001746804,0.000165059,0.000004296471],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004731835,"about_ca_system_score_gemma":0.00002070878,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006741875,"about_ca_topic_score_gemma":0.0001236121,"domain_scores_codex":[0.9990382,0.00002416854,0.0002473769,0.0004088867,0.0001171754,0.000164164],"domain_scores_gemma":[0.9995787,0.00008668813,0.00004478613,0.0002314064,0.00002287214,0.00003553681],"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":[0.000003315314,0.00001776276,0.001488137,0.0002017247,0.00001167151,0.002210362,0.002344832,0.00006324533,0.03550311,0.01190457,0.000195449,0.9460558],"study_design_scores_gemma":[0.0001415032,0.0001397176,0.004259756,0.000445759,0.000005898198,0.002028318,0.0006203009,0.06970789,0.7096655,0.2044425,0.007863584,0.0006792522],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1586301,0.0006043027,0.8379895,0.0001394566,0.0002924105,0.0001052415,1.15474e-7,0.001316558,0.0009223492],"genre_scores_gemma":[0.9151945,0.00004533743,0.08438811,0.00003855556,0.00001771758,0.00001867498,9.01036e-8,0.000007711627,0.0002893543],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9453766,"threshold_uncertainty_score":0.371473,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02030960236271025,"score_gpt":0.3057583953374626,"score_spread":0.2854487929747523,"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."}}