{"id":"W4401070453","doi":"10.1109/jbhi.2024.3435085","title":"CALLM: Enhancing Clinical Interview Analysis Through Data Augmentation With Large Language Models","year":2024,"lang":"en","type":"article","venue":"IEEE Journal of Biomedical and Health Informatics","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"China Scholarship Council","keywords":"Computer science; Natural language processing; Data science; Artificial intelligence","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":[0.003193281,0.001273002,0.0005193561,0.0009720358,0.0004863933,0.0008907363,0.001669946,0.0009864729,0.002849584],"category_scores_gemma":[0.01695275,0.0003891499,0.001059504,0.0006681331,0.0006542933,0.001475144,0.002852282,0.002332388,0.002309396],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00058555,"about_ca_system_score_gemma":0.001616723,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002994034,"about_ca_topic_score_gemma":0.006672261,"domain_scores_codex":[0.9972556,0.001727414,0.00009243904,0.0004500108,0.000365849,0.0001086461],"domain_scores_gemma":[0.993178,0.004634469,0.0002952902,0.0009835791,0.0006960931,0.0002126318],"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":[0.0008841961,0.0008770935,0.008785695,0.0005325435,0.0001804996,0.0005117949,0.001420589,0.1017539,0.03117191,0.005715404,0.0373976,0.8107688],"study_design_scores_gemma":[0.00007598013,0.0003574364,0.002204309,0.00006880263,0.00005675673,0.0003196742,0.0004262392,0.9470249,0.01613164,0.01474879,0.01850143,0.00008414068],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03419657,0.0005008299,0.9487759,0.001528728,0.0003355898,0.0004672287,0.002306466,0.009973789,0.001914745],"genre_scores_gemma":[0.2796257,0.0003205551,0.7031751,0.001332754,0.0002379547,0.001146843,0.0101436,0.0004965465,0.003520928],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003193281,"threshold_uncertainty_score":0.01688784,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1447949056643628,"score_gpt":0.4637682632839094,"score_spread":0.3189733576195466,"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."}}