{"id":"W2401290433","doi":"10.1145/2901739.2901770","title":"Mining duplicate questions in stack overflow","year":2016,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":123,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Computer science; Reputation; Recall rate; Information retrieval; Precision and recall; Data mining; Recall; Stack (abstract data type); Data science; World Wide Web; Artificial intelligence; Programming language","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":[0.0001949272,0.00004933697,0.00005201282,0.0001165917,0.00001687094,0.00004021723,0.000444456,0.00002367815,0.00004393355],"category_scores_gemma":[0.0004431528,0.00003334511,0.00001383319,0.0003058797,0.00001446608,0.0002646533,0.0001674422,0.00004397613,0.0002313631],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005703713,"about_ca_system_score_gemma":0.00003693753,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003170213,"about_ca_topic_score_gemma":0.00001554777,"domain_scores_codex":[0.9993373,0.00001517726,0.00008462307,0.0001949176,0.0001476462,0.0002203358],"domain_scores_gemma":[0.9989495,0.0005555598,0.000007706356,0.0004005059,0.00002650676,0.00006019954],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000005750958,0.000134346,0.4765535,0.00002229421,0.0000165959,0.00009586368,0.0006354225,0.0006924394,0.0182595,0.1777498,0.01268586,0.3131486],"study_design_scores_gemma":[0.0009145844,0.00009775387,0.9318817,0.000160423,9.023721e-7,0.00001985967,0.000009727632,0.03647099,0.01199171,0.003266203,0.01475249,0.0004336794],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1849794,0.00002253554,0.81174,0.001961553,0.00009296086,0.00005279255,3.895598e-7,0.0003204392,0.0008300154],"genre_scores_gemma":[0.9431782,0.000007141581,0.0547498,0.00004027507,0.00001841279,0.00001584612,7.911432e-8,0.000005104675,0.001985107],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7581989,"threshold_uncertainty_score":0.2973782,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02278426483884391,"score_gpt":0.2797278117024878,"score_spread":0.2569435468636439,"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."}}