{"id":"W2108653655","doi":"10.1109/fie.2006.322402","title":"A Tool for Automated GUI Program Grading","year":2006,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of New Brunswick","funders":"","keywords":"Computer science; Java; Graphical user interface; Graphical user interface testing; Software engineering; Grading (engineering); Programming language; Consistency (knowledge bases); Flexibility (engineering); User interface; Operating system; User interface design; Artificial intelligence; Engineering","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":[0.0001865671,0.00006586289,0.00006549177,0.00008189776,0.00005102899,0.0001825728,0.0004559584,0.00002955043,0.000005251285],"category_scores_gemma":[0.0001776949,0.00005773858,0.0000378876,0.000344423,0.0000104815,0.0001149392,0.00008702449,0.00004131832,0.00003003757],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003598604,"about_ca_system_score_gemma":0.00002925594,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003429117,"about_ca_topic_score_gemma":0.000002004487,"domain_scores_codex":[0.9992231,0.000007026576,0.0001035888,0.0001994502,0.0001679583,0.0002988808],"domain_scores_gemma":[0.9993254,0.0002869225,0.00001143814,0.0002813793,0.00006338947,0.0000314696],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000009106153,0.0005501469,0.03220521,0.0001981983,0.00005058856,0.00005005161,0.0001719857,0.007830762,0.0045971,0.3302463,0.2623177,0.3617728],"study_design_scores_gemma":[0.0002236058,0.00007798277,0.01976896,0.000006760547,8.944913e-7,0.000007605489,7.048589e-7,0.9647152,0.002039313,0.001166407,0.01187068,0.0001218222],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.06345167,0.00002216241,0.9258547,0.0002966087,0.0001611233,0.0005422831,6.70891e-7,0.009092481,0.000578236],"genre_scores_gemma":[0.4702221,1.180089e-7,0.52868,0.00001910701,0.00005056011,0.0001803276,0.000001776003,0.000007922245,0.0008380905],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9568845,"threshold_uncertainty_score":0.2354511,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01592801537558914,"score_gpt":0.300303994272294,"score_spread":0.2843759788967049,"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."}}