{"id":"W2105182894","doi":"10.1109/icsmc.2007.4413861","title":"Benchmarking usability of early designs using predictive metrics","year":2007,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Usability; Benchmarking; Computer science; Learnability; Usability inspection; Cognitive walkthrough; Usability goals; Software deployment; Component-based usability testing; Usability engineering; Heuristic evaluation; Software engineering; Human–computer interaction","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":[0.02667765,0.002557453,0.001423741,0.007947372,0.0006216775,0.003681612,0.002232657,0.001322416,0.001180466],"category_scores_gemma":[0.1798739,0.0006964603,0.001231503,0.003195699,0.0009653691,0.005410881,0.001678883,0.001619054,0.0005894037],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002252949,"about_ca_system_score_gemma":0.001344323,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002395778,"about_ca_topic_score_gemma":0.002516516,"domain_scores_codex":[0.9609491,0.0200045,0.002592068,0.002462583,0.01340079,0.0005910146],"domain_scores_gemma":[0.7723507,0.1614875,0.01937193,0.02098306,0.0245324,0.001274496],"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":[0.00117048,0.001636247,0.1806799,0.001513875,0.0007596836,0.0002455128,0.003112343,0.1809792,0.01598715,0.01106308,0.002085937,0.6007665],"study_design_scores_gemma":[0.0001368791,0.004729527,0.09044918,0.0005923675,0.00034204,0.0003639416,0.00100898,0.8529549,0.02663117,0.01741416,0.004971897,0.0004050325],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4555678,0.001066717,0.5341111,0.000280403,0.00007710013,0.0008669704,0.0004505214,0.002233996,0.005345352],"genre_scores_gemma":[0.8220853,0.000306063,0.1752202,0.00005702041,0.00002417149,0.0006669411,0.0008645201,0.0002162422,0.0005595629],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02667765,"threshold_uncertainty_score":0.1410866,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06879968612734985,"score_gpt":0.3205121756188769,"score_spread":0.251712489491527,"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."}}