{"id":"W3048030566","doi":"10.1145/3385732","title":"Automatic Detection of Usability Problem Encounters in Think-aloud Sessions","year":2020,"lang":"en","type":"article","venue":"ACM Transactions on Interactive Intelligent Systems","topic":"Software Engineering Research","field":"Computer Science","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Usability; Think aloud protocol; Computer science; Human–computer interaction; Workflow; Usability lab; Usability inspection; Cognitive walkthrough; Pluralistic walkthrough; Usability engineering; Test (biology); Automation; Usability goals; Engineering; Database","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01631063,0.00209341,0.001253693,0.002790401,0.0009454298,0.002812304,0.001683411,0.001262995,0.002267015],"category_scores_gemma":[0.1089776,0.000899051,0.0005726985,0.001103935,0.0009464614,0.002657707,0.001962283,0.001792942,0.001906077],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000797352,"about_ca_system_score_gemma":0.001309022,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006625039,"about_ca_topic_score_gemma":0.001443937,"domain_scores_codex":[0.9779584,0.01273931,0.001951108,0.002871247,0.003883658,0.0005962774],"domain_scores_gemma":[0.7638327,0.1835862,0.0168623,0.01227748,0.02146241,0.001978907],"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":[0.004325303,0.002991621,0.0775196,0.004317234,0.0003627877,0.0007571304,0.1032255,0.003223283,0.2578245,0.002533701,0.007648355,0.5352709],"study_design_scores_gemma":[0.001146328,0.01108131,0.2965123,0.001981671,0.0006968067,0.002541598,0.05308267,0.1937846,0.3793281,0.02240613,0.03582353,0.001614963],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6922636,0.0003143659,0.2904864,0.0003789366,0.0001582069,0.00438707,0.0009056946,0.00699262,0.004113098],"genre_scores_gemma":[0.7249058,0.0002312136,0.2655666,0.0002938115,0.00007087416,0.004839069,0.001001357,0.0006909299,0.00240038],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01631063,"threshold_uncertainty_score":0.0862599,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03056204144816713,"score_gpt":0.2917840176591617,"score_spread":0.2612219762109945,"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."}}