{"id":"W4406593651","doi":"10.2196/65605","title":"Evaluating the Quality of Psychotherapy Conversational Agents: Framework Development and Cross-Sectional Study","year":2025,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Digital Mental Health Interventions","field":"Psychology","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Quality (philosophy); Psychology; Psychotherapist; Computer science; World Wide Web; Epistemology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1462776,0.0008461294,0.001181126,0.005961826,0.002625859,0.0027247,0.001458908,0.0009719868,0.002006232],"category_scores_gemma":[0.1343809,0.001060394,0.002762982,0.003910845,0.002487849,0.003074173,0.004169223,0.001435577,0.0003041757],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006952656,"about_ca_system_score_gemma":0.007626434,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01104594,"about_ca_topic_score_gemma":0.01919115,"domain_scores_codex":[0.9283177,0.05426579,0.0055123,0.00412127,0.006279199,0.00150376],"domain_scores_gemma":[0.8190865,0.1114679,0.01962242,0.00970643,0.03708103,0.003035861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005439849,0.001534155,0.9243861,0.00104661,0.0008478942,0.0001433293,0.03191917,0.000528501,0.0003710787,0.001954881,0.001221561,0.03550278],"study_design_scores_gemma":[0.0004814798,0.007332091,0.9214914,0.002135435,0.001707209,0.0005577304,0.04320891,0.009784394,0.001756513,0.00177653,0.009591099,0.0001772883],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9600534,0.001356148,0.0229577,0.000307619,0.00005640763,0.01203991,0.000859836,0.00008418252,0.002284872],"genre_scores_gemma":[0.9311232,0.0005015323,0.04747387,0.0001612157,0.00002450875,0.01940388,0.0009156244,0.00003718396,0.0003589588],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1462776,"threshold_uncertainty_score":0.7735991,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4125547857148369,"score_gpt":0.6713810563273551,"score_spread":0.2588262706125182,"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."}}