Vanin Alexander, Bolshev Vadim, Panfilova Anastasia
Laboratory of AI Technologies in Psychology, Institute of Psychology Russian Academy of Sciences, Moscow, Russia.
Front Psychiatry. 2025 Jul 25;16:1608163. doi: 10.3389/fpsyt.2025.1608163. eCollection 2025.
This paper addresses the growing intersection of machine learning (ML) and psychotherapy by developing a classification model for analyzing topics in therapist remarks. Understanding recurring language patterns in therapist communication can enhance clinical practice, supervision, and training, yet systematic approaches to topic analysis remain limited.
The study applies BERTopic, an ML-based topic modeling technique, to unstructured dialogues from two distinct groups of therapists: classical (founders of therapeutic schools such as Carl Rogers, Fritz Perls, and Albert Ellis) and modern practitioners representing diverse psychotherapeutic approaches. The implementation involves constructing a vector space of therapist remarks, applying dimensionality reduction, clustering, and optimizing topic representations. To improve interpretability, expert assessment and manual refinement complement the automated modeling process. The resulting topics are used as features to train an ML classifier, which is then tested on a case study comparing Carl Rogers' sessions with those of modern Cognitive Behavioral Therapy (CBT) practitioners.
The analysis identifies the most common and stable topics across both therapist groups, highlighting recurring patterns and unique thematic compositions. The case study reveals distinct differences in thematic structures, with key topics emerging that characterize each group's therapeutic discourse. The trained classifier demonstrates robust performance in distinguishing these thematic patterns.
The study shows that automated topic modeling, combined with expert input, can effectively uncover how therapist language patterns emerge and persist across different therapeutic styles. The resulting model, made publicly available, offers broad applications in psychotherapy research, clinical supervision, and training. These findings underscore the potential of topic modeling as a valuable tool for deepening our understanding of therapist communication and advancing ML applications in psychotherapy.
本文通过开发一种用于分析治疗师言论主题的分类模型，探讨了机器学习（ML）与心理治疗日益增长的交叉领域。理解治疗师沟通中反复出现的语言模式可以加强临床实践、督导和培训，但主题分析的系统方法仍然有限。
本研究将基于机器学习的主题建模技术BERTopic应用于两组不同治疗师的非结构化对话：经典治疗师（如卡尔·罗杰斯、弗里茨·皮尔斯和阿尔伯特·艾利斯等治疗学派的创始人）以及代表不同心理治疗方法的现代从业者。实施过程包括构建治疗师言论的向量空间、应用降维、聚类以及优化主题表示。为提高可解释性，专家评估和人工细化对自动化建模过程起到补充作用。所得主题用作特征来训练一个机器学习分类器，然后在一个案例研究中进行测试，该案例研究比较了卡尔·罗杰斯的治疗 sessions 与现代认知行为疗法（CBT）从业者的 sessions。
分析确定了两组治疗师中最常见和稳定的主题，突出了反复出现的模式和独特的主题构成。案例研究揭示了主题结构上的明显差异，出现了表征每组治疗话语的关键主题。经过训练的分类器在区分这些主题模式方面表现出强大的性能。
该研究表明，自动化主题建模与专家输入相结合，可以有效地揭示治疗师语言模式如何在不同治疗风格中出现和持续存在。所得模型已公开提供，在心理治疗研究、临床督导和培训中有广泛应用。这些发现强调了主题建模作为一种有价值工具的潜力，可加深我们对治疗师沟通的理解，并推动机器学习在心理治疗中的应用。
