{"id":"W3208243843","doi":"10.1109/rew53955.2021.00025","title":"Is BERT the New Silver Bullet? - An Empirical Investigation of Requirements Dependency Classification","year":2021,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Computer science; Transformer; Artificial intelligence; Silver bullet; Dependency (UML); Empirical research; Encoder; Machine learning; Data mining; Engineering","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":[0.0179953,0.0005442234,0.0004679122,0.001848762,0.0006685978,0.001417314,0.001177473,0.001065389,0.002105535],"category_scores_gemma":[0.1161931,0.0003172304,0.0006594738,0.002251955,0.001346482,0.004820677,0.001391272,0.002973429,0.0007772667],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001077288,"about_ca_system_score_gemma":0.000615843,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003115072,"about_ca_topic_score_gemma":0.005857886,"domain_scores_codex":[0.989907,0.006138241,0.0007394235,0.001175564,0.001741988,0.0002978796],"domain_scores_gemma":[0.7763771,0.1974465,0.007437137,0.01266936,0.005164891,0.0009050024],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002248147,0.002725724,0.5123225,0.001036851,0.0003957541,0.0006933938,0.004092393,0.0654376,0.01046412,0.01318516,0.01929044,0.3681079],"study_design_scores_gemma":[0.0002319334,0.001348222,0.2561051,0.0003977044,0.0001678383,0.001633391,0.004737504,0.6671355,0.01733237,0.028323,0.02243579,0.0001516418],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9710811,0.0005160649,0.02354411,0.0007892976,0.00002785512,0.0001203137,0.001298554,0.0002860564,0.002336675],"genre_scores_gemma":[0.9813757,0.0001019218,0.01496236,0.0001502895,0.00001955871,0.00009058823,0.00277648,0.00007967433,0.0004434652],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0179953,"threshold_uncertainty_score":0.09516937,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1038612302168992,"score_gpt":0.3512246052520377,"score_spread":0.2473633750351385,"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."}}