{"id":"W2341384651","doi":"10.1145/2911451.2911510","title":"Engineering Quality and Reliability in Technology-Assisted Review","year":2016,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":96,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Reliability (semiconductor); Reliability engineering; Quality (philosophy); Computer science; Measure (data warehouse); Software quality; Taguchi methods; Data mining; Machine learning; Engineering; Software","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":[0.1171493,0.001381183,0.002171167,0.007760001,0.001356683,0.006979066,0.002711179,0.002287258,0.001892586],"category_scores_gemma":[0.4840193,0.001503375,0.001456285,0.005476467,0.003175644,0.007708445,0.004027444,0.001944577,0.001131867],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004114216,"about_ca_system_score_gemma":0.005472884,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003285334,"about_ca_topic_score_gemma":0.004005957,"domain_scores_codex":[0.7842729,0.1503779,0.01387536,0.009772909,0.03992428,0.001776707],"domain_scores_gemma":[0.42393,0.4128564,0.04586503,0.04981492,0.06539743,0.002136271],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0008916711,0.0002699376,0.02833357,0.005437699,0.0009037573,0.0004447681,0.003752085,0.1248894,0.01332193,0.08143084,0.009122916,0.7312014],"study_design_scores_gemma":[0.0005450579,0.002933096,0.04512767,0.001998643,0.0009726007,0.002199323,0.001782431,0.5588883,0.03638415,0.2961436,0.05237874,0.000646453],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04569474,0.00647238,0.9345844,0.003168295,0.0002084118,0.0008700797,0.0002594347,0.00144731,0.007294901],"genre_scores_gemma":[0.5582738,0.002139611,0.4342692,0.0006663995,0.0003528669,0.001068509,0.0004287679,0.0004528726,0.002348029],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1171493,"threshold_uncertainty_score":0.6195522,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02710917404402359,"score_gpt":0.3098347032712329,"score_spread":0.2827255292272093,"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."}}