{"id":"W3014728464","doi":"10.1109/saner48275.2020.9054848","title":"Studying Developer Reading Behavior on Stack Overflow during API Summarization Tasks","year":2020,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"École de Technologie Supérieure; Carleton University","funders":"","keywords":"Computer science; Automatic summarization; Source code; World Wide Web; Reading (process); Software; Program comprehension; Gaze; Reading comprehension; Code (set theory); Information retrieval; Artificial intelligence; Programming language; Software system","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":[0.0001584223,0.0001304095,0.0001165621,0.0001088315,0.0001252987,0.000231133,0.0006029769,0.00004609525,0.00005354969],"category_scores_gemma":[0.0005151846,0.0001249652,0.00002830274,0.0006296419,0.000009893067,0.0004232271,0.0003804065,0.0002045473,0.0001797211],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001104036,"about_ca_system_score_gemma":0.00005310814,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001454803,"about_ca_topic_score_gemma":0.000001177722,"domain_scores_codex":[0.9985734,0.00003027632,0.0001687032,0.0004249169,0.0004806775,0.0003219939],"domain_scores_gemma":[0.9992176,0.0001924013,0.00002382812,0.0003346776,0.00007261201,0.0001588283],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005564966,0.0003191708,0.8110408,0.0002512344,0.0001007038,0.0005314043,0.009960539,0.02073886,0.0703209,0.01406405,0.003409785,0.06920691],"study_design_scores_gemma":[0.001075338,0.0002582128,0.8282744,0.00009561663,0.00001057042,0.00001321837,0.0001162765,0.07897718,0.08906466,0.00002904837,0.001279644,0.0008058703],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6415766,0.00000910438,0.3564419,0.0004817372,0.0001588198,0.0002277787,9.70899e-7,0.0007049913,0.0003981933],"genre_scores_gemma":[0.9758252,0.000003903531,0.02342603,0.0002118937,0.00007466684,0.00002993935,0.000002822985,0.00002141026,0.0004041391],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3342486,"threshold_uncertainty_score":0.5095932,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04476517789639843,"score_gpt":0.2741167396505266,"score_spread":0.2293515617541281,"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."}}