{"id":"W4393548574","doi":"10.5281/zenodo.4010209","title":"Conclusion Stability for Natural Language Based Mining of Design Discussions: Dataset","year":2020,"lang":"en","type":"dataset","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 Victoria","funders":"","keywords":"Natural (archaeology); Stability (learning theory); Computer science; Natural language processing; Geography; Archaeology; Machine learning","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":[0.001556299,0.002443064,0.0008153648,0.003813985,0.0009672919,0.001788402,0.003095758,0.002462263,0.02472429],"category_scores_gemma":[0.007298565,0.0005336661,0.001539846,0.003229665,0.000515966,0.001079332,0.001698155,0.001700106,0.03420071],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001573947,"about_ca_system_score_gemma":0.002488205,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01077364,"about_ca_topic_score_gemma":0.01997781,"domain_scores_codex":[0.9981401,0.0003064038,0.0002082736,0.0005430526,0.0006031455,0.0001990124],"domain_scores_gemma":[0.9958227,0.00129351,0.0004281772,0.001196546,0.0009331204,0.000325965],"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":[0.0003428442,0.000213266,0.004890903,0.001760689,0.00008276311,0.00009179774,0.0001116149,0.001902249,0.000890232,0.001297808,0.9747677,0.01364803],"study_design_scores_gemma":[0.0006790787,0.0002002036,0.02292413,0.0004924317,0.0001191462,0.000288325,0.0003327882,0.006613417,0.004490438,0.003329527,0.9604337,0.00009684722],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.00256689,0.0001095234,0.0006461876,0.00009996872,0.00003296361,0.00007715641,0.9935154,0.001515837,0.001436055],"genre_scores_gemma":[0.002408253,0.00003840102,0.001521933,0.00004535768,0.000007211466,0.0002337058,0.9948471,0.00008079191,0.0008172823],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.02472429,"threshold_uncertainty_score":0.08271104,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0511869501477251,"score_gpt":0.2814128574786919,"score_spread":0.2302259073309668,"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."}}