{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01222398,0.0001118657,0.0003871885,0.0003879385,0.00008178156,0.00002391138,0.0002864747,0.0002146012,0.00008971703],"category_scores_gemma":[0.004250073,0.00008385866,0.00009976325,0.0005295334,0.0001730739,0.001791698,0.00003933662,0.0002061662,0.000002946283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008427442,"about_ca_system_score_gemma":0.0002733454,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000112649,"about_ca_topic_score_gemma":0.000003121246,"domain_scores_codex":[0.9957565,0.0006553709,0.001379274,0.0001736103,0.001882054,0.000153187],"domain_scores_gemma":[0.9933156,0.0003316358,0.001605809,0.0002593821,0.004342165,0.0001454028],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000142177,0.0005269755,0.9707064,0.00003717551,0.00001814484,1.960691e-7,0.002273011,0.002978278,0.0006276515,0.0001958488,0.000137757,0.02235638],"study_design_scores_gemma":[0.000922201,0.0001439271,0.9291192,0.00004994084,0.00008817414,0.00004107597,0.001317771,0.0382297,0.001892183,0.02807654,0.00001832852,0.000100997],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9215658,0.0004105807,0.07747798,0.0001228531,0.0001983907,0.0001304626,0.00002342352,0.00005666111,0.00001379594],"genre_scores_gemma":[0.9950716,0.000006972057,0.00484061,0.00002081028,0.00003593372,0.000003028571,0.000004312035,0.000006025661,0.00001064322],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07350581,"threshold_uncertainty_score":0.5088042,"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."}}