{"id":"W3122939027","doi":"10.7287/peerj.preprints.1705","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":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Crash; Computer science; Data deduplication; Software; Scalability; Information retrieval; Precision and recall; Set (abstract data type); Database; Data science; Data mining; Software engineering; World Wide Web; 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":[0.01226868,0.001404608,0.001919828,0.005956532,0.001853453,0.004512848,0.003199164,0.00212774,0.001740925],"category_scores_gemma":[0.05541085,0.0006992496,0.0009299394,0.00689789,0.001708833,0.01070256,0.002119615,0.001726293,0.003701755],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001393293,"about_ca_system_score_gemma":0.002206867,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004745031,"about_ca_topic_score_gemma":0.006238979,"domain_scores_codex":[0.9896402,0.002831167,0.001293832,0.001752099,0.003981525,0.0005011604],"domain_scores_gemma":[0.9397359,0.03391036,0.002765676,0.01678597,0.006293709,0.0005083526],"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":[0.001388871,0.0005877083,0.01464553,0.002703605,0.0004481426,0.0004463361,0.001064987,0.02243709,0.04002628,0.004833017,0.05123869,0.8601797],"study_design_scores_gemma":[0.0007431934,0.003742386,0.04628687,0.001059934,0.001165324,0.008314438,0.005449886,0.3765667,0.3769151,0.03763249,0.1413299,0.0007937249],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5317281,0.04095521,0.3574507,0.006870482,0.002093017,0.001539051,0.00619617,0.0314094,0.02175775],"genre_scores_gemma":[0.6811616,0.006312379,0.2966744,0.001348951,0.0005866133,0.0002932742,0.005778881,0.0009282769,0.006915636],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01226868,"threshold_uncertainty_score":0.06488377,"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."}}