The usefulness of combining topic modelling and statistical analysis to investigate the therapeutic process: A single case study.

Abstract

Objective: This study examines whether patterns in the movement of topics during psychotherapy sessions can provide psychotherapists with actionable insights for single-case analysis. It utilizes both statistical models and AI-driven tools to uncover these dynamics.

Method: We transcribed a completed psychotherapy session comprising 26 sessions. First, common topics across all therapies were identified, and then expert psychotherapists labelled each conversational turn of this selected psychotherapy. As determined by the experts, the topic dynamics were analysed using Generalized Additive Mixed Models (GAMMs), which captured non-linear trends and hierarchical structures within the data. Subsequently, these trajectories, as identified by the experts, were compared with the topics extracted in an unsupervised manner using a topic modelling algorithm, called Latent Dirichlet Allocation (LDA).

Results: Our findings confirm that topic trajectory analysis reliably indicates therapeutic progress. Specifically, topics related to suffering (SPS) decreased over time, while topics concerning therapeutic refiguration and insight (TRI) increased, reflecting clinical improvement.

Conclusion: The study demonstrates that both GAMMs and LDA are useful tools to see how the topics in specific psychotherapy are modified their occurrence during the therapeutic work. Combining classical methods of statistical analysis and AI-driven topic analysis enhances the sensitivity of assessments, providing insights into how the psychotherapy work changes across sessions.