Leveraging natural language processing to enhance feedback-informed group therapy: A proof of concept.

Group therapy has evolved as a powerful therapeutic approach, facilitating mutual support, interpersonal learning, and personal growth among members. However, the complexity of studying communication dynamics, emotional expressions, and group interactions between multiple members and often coleaders is a frequent barrier to advancing group therapy research and practice. Fortunately, advances in machine learning technologies, for example, natural language processing (NLP), make it possible to study these complex verbal and behavioral interactions within a small group. Additionally, these technologies may serve to provide leaders and members with important and actionable feedback about group therapy sessions, possibly enhancing the utility of feedback-informed care in group therapy. As such, this study sought to provide a proof of concept for applying NLP technologies to automatically assess alliance ratings from participant utterances in two community-based online support groups for weight stigma. We compared traditional machine learning approaches with advanced transformer-based language models, including variants pretrained on mental health and psychotherapy data. Results indicated that several models detected relationships between participant utterances and alliance, with the best performing model achieving an area under the receiver operating characteristic curve of 0.654. Logistic regression analysis identified specific utterances associated with high and low alliance ratings, providing interpretable insights into group dynamics. While acknowledging limitations such as small sample size and the specific context of weight stigma groups, this study provides insights into the potential of NLP in augmenting feedback-informed group therapy. Implications for real-time process monitoring and future directions for enhancing model performance in diverse group therapy settings are discussed. (PsycInfo Database Record (c) 2025 APA, all rights reserved).