{"id":"W2161466705","doi":"10.1002/smr.1562","title":"Automated evaluation of website navigability: an empirical validation of multilevel quality models","year":2012,"lang":"en","type":"article","venue":"Journal of Software Evolution and Process","topic":"Technology Adoption and User Behaviour","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Université du Québec à Montréal","funders":"","keywords":"Navigability; Quality (philosophy); Computer science; Test (biology); Empirical research; Information retrieval; Data science; Statistics; Mathematics","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":[0.02303397,0.0007012061,0.0007769727,0.002549059,0.0005407953,0.003027384,0.001549981,0.001387947,0.001798373],"category_scores_gemma":[0.1384495,0.0005159265,0.001444871,0.002503198,0.001089755,0.003127575,0.002112461,0.001834115,0.0004411007],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002121516,"about_ca_system_score_gemma":0.0009400491,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008106124,"about_ca_topic_score_gemma":0.006112813,"domain_scores_codex":[0.9807379,0.01352447,0.001004222,0.002030897,0.002261632,0.000440951],"domain_scores_gemma":[0.6944931,0.2410562,0.02690061,0.02037566,0.01489371,0.002280753],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001773169,0.002789673,0.8562089,0.0002685798,0.0008001658,0.0000895737,0.002388954,0.06904946,0.001622399,0.002756985,0.001146788,0.06110528],"study_design_scores_gemma":[0.0001211461,0.001222364,0.1998859,0.00005722826,0.0001736982,0.00005368351,0.0004573609,0.7932851,0.001323731,0.003042666,0.0003082103,0.00006884388],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9796464,0.00006057759,0.01882996,0.00009062009,0.000008529022,0.0001351194,0.0002436566,0.0001609275,0.0008242473],"genre_scores_gemma":[0.9890933,0.00001473758,0.01046863,0.0000102877,0.000004423698,0.0000851907,0.0002188714,0.00001712712,0.00008734661],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02303397,"threshold_uncertainty_score":0.1218167,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3180004297024163,"score_gpt":0.5154014539467916,"score_spread":0.1974010242443753,"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."}}