{"id":"W6949929472","doi":"10.5281/zenodo.3839074","title":"Characterizing Task-Relevant Information in Natural Language Software Artifacts","year":2020,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Artifact (error); Natural language; Task (project management); Consistency (knowledge bases); Software; Identification (biology); Set (abstract data type); Software development","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":[0.003176958,0.0006242347,0.0002922355,0.002370937,0.0007328728,0.001945125,0.000727427,0.0009426285,0.02765811],"category_scores_gemma":[0.04682869,0.0002282097,0.0003962458,0.001496769,0.0003951498,0.002019792,0.001327594,0.000661513,0.006423256],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007457485,"about_ca_system_score_gemma":0.0010123,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001842834,"about_ca_topic_score_gemma":0.004128514,"domain_scores_codex":[0.9973282,0.001133529,0.0002836291,0.0005620093,0.0005861935,0.0001063132],"domain_scores_gemma":[0.9220001,0.0629781,0.003778341,0.003493468,0.007019519,0.0007306067],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002094053,0.002347941,0.05717227,0.009763328,0.0001970292,0.00201504,0.009998286,0.006433312,0.1112838,0.02069013,0.2721895,0.5058154],"study_design_scores_gemma":[0.0005787687,0.001933508,0.2384707,0.002577866,0.0003608122,0.004463226,0.01078517,0.1323086,0.1222597,0.05639923,0.4292893,0.0005731949],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5025124,0.001886932,0.3207963,0.003173346,0.0009137184,0.002677332,0.1229301,0.01036134,0.03474849],"genre_scores_gemma":[0.5224282,0.0006311604,0.3155859,0.0009654378,0.0002581775,0.002611482,0.1471693,0.001328105,0.009022213],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02765811,"threshold_uncertainty_score":0.0925256,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01812008366101989,"score_gpt":0.2233158210150097,"score_spread":0.2051957373539899,"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."}}