Employing large language models for emotion detection in psychotherapy transcripts
Frontiers in Psychiatry · 09.05.2025 · 16
DOI: 10.3389/fpsyt.2025.1504306 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Psychiatry, online erschienen am 09.05.2025. Verfasst von Christopher Lalk und 8 weiteren Autor:innen. Beteiligte Einrichtung: Department of Psychology, Osnabrück University, Osnabrück, Germany. Die Arbeit wurde bisher 20-mal zitiert (Crossref) und nennt 98 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Deutsche Forschungsgemeinschaft.
Abstract (Originalsprache)
Purpose
In the context of psychotherapy, emotions play an important role both through their association with symptom severity, as well as their effects on the therapeutic relationship. In this analysis, we aim to train a large language model (LLM) for the detection of emotions in German speech. We want to apply this model on a corpus of psychotherapy transcripts to predict symptom severity and alliance aiming to identify the most important emotions for the prediction of symptom severity and therapeutic alliance.
Methods
We employed a public labeled dataset of 28 emotions and translated the dataset into German. A pre-trained LLM was then fine-tuned on this dataset for emotion classification. We applied the fine-tuned model to a dataset containing 553 psychotherapy sessions of 124 patients. Using machine learning (ML) and explainable artificial intelligence (AI), we predicted symptom severity and alliance by the detected emotions.
Results
Our fine-tuned model achieved modest classification performance ( F1 macro =0.45, Accuracy =0.41, Kappa =0.42) across the 28 emotions. Incorporating all emotions, our ML model showed satisfying performance for the prediction of symptom severity ( r = .50; 95%-CI:.42,.57) and moderate performance for the prediction of alliance scores ( r = .20; 95%-CI:.06,.32). The most important emotions for the prediction of symptom severity were approval, anger , and fear . The most important emotions for the prediction of alliance were curiosity, confusion , and surprise .
Conclusions
Even though the classification results were only moderate, our model achieved a good performance especially for prediction of symptom severity. The results confirm the role of negative emotions in the prediction of symptom severity, while they also highlight the role of positive emotions in fostering a good alliance. Future directions entail the improvement of the labeled dataset, especially with regards to domain-specificity and incorporating context information. Additionally, other modalities and Natural Language Processsing (NLP)-based alliance assessment could be integrated.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Psychology, Osnabrück University, Osnabrück, Germany.
- Department of Psychology, University of Trier, Trier, Germany.
Förderung
- Deutsche Forschungsgemeinschaft
Zitieren
APA
Lalk, C., Targan, K., Steinbrenner, T., Schaffrath, J., Eberhardt, S., Schwartz, B., Vehlen, A., Lutz, W., & Rubel, J. (2025). Employing large language models for emotion detection in psychotherapy transcripts. Frontiers in Psychiatry, 16. https://doi.org/10.3389/fpsyt.2025.1504306
Vancouver
Lalk C, Targan K, Steinbrenner T, Schaffrath J, Eberhardt S, Schwartz B, et al. Employing large language models for emotion detection in psychotherapy transcripts. Frontiers in Psychiatry. 2025;16. doi:10.3389/fpsyt.2025.1504306
BibTeX
@article{Lalk2025,
author = {Lalk, Christopher and Targan, Kim and Steinbrenner, Tobias and Schaffrath, Jana and Eberhardt, Steffen and Schwartz, Brian and Vehlen, Antonia and Lutz, Wolfgang and Rubel, Julian},
title = {Employing large language models for emotion detection in psychotherapy transcripts},
journal = {Frontiers in Psychiatry},
year = {2025},
volume = {16},
doi = {10.3389/fpsyt.2025.1504306}
}
EUVIMED