{"id":"W2258726425","doi":"10.1145/2901739.2901766","title":"The unreasonable effectiveness of traditional information retrieval in crash report deduplication","year":2016,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Mozilla Foundation; Mitacs","keywords":"Crash; Computer science; Data deduplication; Software; Scalability; Information retrieval; Precision and recall; Database; Data science; Software engineering; World Wide Web; Operating system","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":[0.00153136,0.00003874696,0.00005090508,0.00007425903,0.00003244558,0.0000340703,0.0003217003,0.00002691452,0.00000473617],"category_scores_gemma":[0.002021726,0.00002177597,0.00001808998,0.0003709373,0.00003024497,0.000761123,0.00005178107,0.00004083009,0.00001774481],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009759289,"about_ca_system_score_gemma":0.0001035724,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000162101,"about_ca_topic_score_gemma":0.000001360503,"domain_scores_codex":[0.9992725,0.00004917267,0.000172889,0.0000893605,0.0003006832,0.0001153894],"domain_scores_gemma":[0.9965084,0.002978822,0.00004401887,0.0003082741,0.0001354239,0.00002502371],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0001217406,0.0001396608,0.1757207,0.0001459912,0.00003089126,0.00001621558,0.0001811225,0.0007646575,0.01689698,0.7107293,0.001470736,0.09378207],"study_design_scores_gemma":[0.0002559987,0.00003225257,0.9613242,0.00004349548,3.815732e-7,0.00003024731,0.000002130051,0.002588583,0.02776295,0.007186848,0.0007247716,0.00004816557],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2827084,0.00001022586,0.7159863,0.0004086835,0.0001044941,0.0001595754,9.529439e-7,0.0000643244,0.000557048],"genre_scores_gemma":[0.9981321,0.00000504741,0.001755718,0.000005527947,0.000009598008,0.00002119579,0.000002252111,0.000001759004,0.0000668346],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7856035,"threshold_uncertainty_score":0.2420341,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01526597073158071,"score_gpt":0.2492900997646688,"score_spread":0.2340241290330881,"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."}}