{"id":"W4240162129","doi":"10.31234/osf.io/wmyuc","title":"A Systematic Review of Machine Learning for Assessment and Feedback of Treatment Fidelity","year":2021,"lang":"en","type":"review","venue":"","topic":"Digital Mental Health Interventions","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Fidelity; Computer science; Machine learning; Coding (social sciences); Implementation; Artificial intelligence; Scalability; Session (web analytics); Code review; Utterance; Software quality; Software engineering; Database; Software; World Wide Web","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04768862,0.001949703,0.008670668,0.01742056,0.001146337,0.00405992,0.00337395,0.00237798,0.005907218],"category_scores_gemma":[0.2093345,0.001537573,0.007189089,0.01286001,0.001700599,0.003997653,0.002772443,0.002293137,0.0007503887],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007805579,"about_ca_system_score_gemma":0.02373577,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00891957,"about_ca_topic_score_gemma":0.02576504,"domain_scores_codex":[0.9549314,0.02011506,0.01513487,0.001586556,0.007864411,0.0003678343],"domain_scores_gemma":[0.820438,0.1442159,0.01693481,0.003640221,0.01407447,0.0006966558],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0001411182,0.00002189457,0.0003565842,0.7962977,0.004629334,0.00005630527,0.0003000034,0.0002177148,0.0001156186,0.0008832196,0.004830641,0.1921498],"study_design_scores_gemma":[0.0001117155,0.00008879571,0.001000662,0.9594426,0.01055097,0.0001295184,0.0001149043,0.0001492389,0.0001370891,0.0008199876,0.02742033,0.00003418043],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002936932,0.9959874,0.001167692,0.0007735665,0.000228358,0.0007492639,0.0003454062,0.00002570498,0.0004288192],"genre_scores_gemma":[0.006138225,0.9852658,0.005240987,0.000770474,0.0001146191,0.001916763,0.0003606807,0.00002174213,0.0001707573],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9523114,"threshold_uncertainty_score":0.2522045,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1647486907061495,"score_gpt":0.5231395466785178,"score_spread":0.3583908559723683,"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."}}