{"id":"W4415125743","doi":"10.1109/rew66121.2025.00014","title":"Automatic Classification of User Requirements from Online Feedback - A Replication Study","year":2025,"lang":"en","type":"article","venue":"","topic":"Big Data and Business Intelligence","field":"Business, Management and Accounting","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"University of Calgary","keywords":"Baseline (sea); Replication (statistics); Workflow; Replicate; Deep learning; Ambiguity","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01366134,0.001565574,0.001145813,0.001521136,0.0006938286,0.001773006,0.002318207,0.001812257,0.001600632],"category_scores_gemma":[0.04681609,0.0005364451,0.001952374,0.0009645283,0.00104854,0.003109301,0.001712646,0.002905099,0.002317227],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001145435,"about_ca_system_score_gemma":0.001217351,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006355832,"about_ca_topic_score_gemma":0.006023721,"domain_scores_codex":[0.9879935,0.005982738,0.0009986784,0.002878659,0.001828033,0.0003183996],"domain_scores_gemma":[0.9620391,0.01272408,0.001344406,0.01614745,0.007041546,0.000703319],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.006533738,0.01232757,0.1515537,0.003343894,0.001897867,0.001506715,0.004167209,0.04223116,0.04451978,0.002645805,0.05347481,0.6757978],"study_design_scores_gemma":[0.002872378,0.01321293,0.2217562,0.0007223569,0.001414867,0.002521921,0.005454573,0.5955307,0.08858383,0.008739123,0.05850068,0.0006905124],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9319252,0.001454729,0.04718106,0.001048544,0.0007636608,0.002179333,0.005328322,0.005089625,0.005029497],"genre_scores_gemma":[0.917609,0.0003673671,0.06300847,0.0006987294,0.0001901402,0.001368773,0.01296707,0.0003179988,0.003472542],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9863387,"threshold_uncertainty_score":0.07224894,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1288946041252902,"score_gpt":0.3592811471389075,"score_spread":0.2303865430136173,"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."}}