{"id":"W4384026505","doi":"10.1109/msr59073.2023.00067","title":"DACOS—A Manually Annotated Dataset of Code Smells","year":2023,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Computer science; Code smell; Code (set theory); Context (archaeology); Information retrieval; Benchmarking; Artificial intelligence; Focus (optics); Machine learning; Natural language processing; World Wide Web; 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":[0.00160059,0.001564634,0.0007187729,0.008043788,0.001110516,0.001098062,0.001405513,0.001819432,0.003751473],"category_scores_gemma":[0.01304204,0.0004652241,0.0008188316,0.006199817,0.0007296469,0.001934353,0.001573087,0.001516429,0.005075737],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001189171,"about_ca_system_score_gemma":0.001892583,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01060024,"about_ca_topic_score_gemma":0.02887234,"domain_scores_codex":[0.9964153,0.000483726,0.0005334481,0.0009185991,0.001419164,0.0002297202],"domain_scores_gemma":[0.9831759,0.004652188,0.002347723,0.00297329,0.006068885,0.0007820598],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00170907,0.0006660691,0.05079519,0.006160488,0.0002448626,0.001615255,0.001951899,0.006930387,0.03765341,0.003460489,0.7138473,0.1749656],"study_design_scores_gemma":[0.0003126944,0.0005713145,0.1363745,0.0009724617,0.0001535133,0.001860455,0.001401381,0.0355271,0.03911233,0.00467372,0.7786646,0.0003760396],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.103848,0.001875042,0.01576447,0.0005733268,0.0003714518,0.0004406045,0.8450893,0.02312479,0.008913101],"genre_scores_gemma":[0.04619895,0.0004165512,0.02676593,0.0001884816,0.00005236781,0.0005617162,0.9210564,0.001367002,0.003392595],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.01060024,"threshold_uncertainty_score":0.0210771,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03287142682747221,"score_gpt":0.3131079492889892,"score_spread":0.280236522461517,"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."}}