{"id":"W2883027691","doi":"10.1145/3196398.3196457","title":"Developer interaction traces backed by IDE screen recordings from think aloud sessions","year":2018,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal; Concordia University","funders":"Polytechnique Montréal","keywords":"Correctness; Computer science; Program comprehension; Set (abstract data type); Human–computer interaction; Empirical research; Data science; Comprehension; Ground truth; Artificial intelligence; Software; Programming language; Software system","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":[0.001556358,0.0007049445,0.0004743153,0.005374329,0.000557525,0.001009663,0.0005709768,0.0007786049,0.003310053],"category_scores_gemma":[0.0214619,0.000351062,0.0002462695,0.004474657,0.0005388632,0.00111409,0.001185744,0.001194472,0.002313791],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003632093,"about_ca_system_score_gemma":0.0005975894,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00371751,"about_ca_topic_score_gemma":0.008474939,"domain_scores_codex":[0.9975998,0.0006142561,0.0002542671,0.0003749648,0.0009794086,0.0001772856],"domain_scores_gemma":[0.9684542,0.01772867,0.002975708,0.00410678,0.005749889,0.0009847468],"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":[0.001963227,0.001877478,0.2504823,0.004302089,0.0003075223,0.00270279,0.0534493,0.004114395,0.0560157,0.004252459,0.07778579,0.542747],"study_design_scores_gemma":[0.0001680829,0.001248225,0.7774875,0.001096586,0.0001268794,0.002006678,0.02784134,0.01730241,0.04390669,0.007454375,0.1211044,0.0002569012],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8355318,0.0006110426,0.05735648,0.000603292,0.0002035362,0.0009529534,0.08578837,0.003385516,0.01556709],"genre_scores_gemma":[0.8610653,0.0005338806,0.06307811,0.0001262014,0.0001214983,0.00168404,0.06683143,0.0006065342,0.005953007],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005374329,"threshold_uncertainty_score":0.01107323,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02411048900598236,"score_gpt":0.2914301773040616,"score_spread":0.2673196882980792,"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."}}