{"id":"W2069169007","doi":"10.1109/wse.2012.6320526","title":"Visual testing of Graphical User Interfaces: An exploratory study towards systematic definitions and approaches","year":2012,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Graphical user interface; Computer science; Graphical user interface testing; Human–computer interaction; User interface; Visualization; User interface design; User experience design; Artificial intelligence; Programming language","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":[],"consensus_categories":[],"category_scores_codex":[0.03867522,0.0008796976,0.0008285689,0.007470798,0.002113699,0.004485149,0.002760137,0.001642663,0.0008141059],"category_scores_gemma":[0.1373564,0.0007978407,0.0006110286,0.004898458,0.008489107,0.007358208,0.005952691,0.002440766,0.0001511859],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00373522,"about_ca_system_score_gemma":0.0070116,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002986981,"about_ca_topic_score_gemma":0.004284854,"domain_scores_codex":[0.9251545,0.05964667,0.003832554,0.002374541,0.007812857,0.001178893],"domain_scores_gemma":[0.6916152,0.2606469,0.01736611,0.01222075,0.01656983,0.001581127],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0002522316,0.001437596,0.07200313,0.005005461,0.00007663319,0.001451161,0.4677989,0.001605386,0.01272294,0.03583344,0.001117921,0.4006953],"study_design_scores_gemma":[0.0002694999,0.004351214,0.1238849,0.02232224,0.000276058,0.006821616,0.647072,0.01349617,0.02816086,0.054611,0.09841506,0.0003193945],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8030003,0.01145738,0.1666871,0.002761069,0.0000478816,0.002560529,0.0001925867,0.00020562,0.01308749],"genre_scores_gemma":[0.8988955,0.004285885,0.09379148,0.0006029297,0.00001958521,0.001275998,0.0001642807,0.00008256236,0.0008817853],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03867522,"threshold_uncertainty_score":0.2045366,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2284178662324735,"score_gpt":0.3159080840569035,"score_spread":0.08749021782443,"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."}}