DMT-CBT: Longitudinal Therapeutic State Modeling for CBT Counseling

2026-06-02

AI Summary

This work addresses the limitation of existing large language models in cognitive behavioral therapy (CBT) counseling, which typically focus on single-turn, short-text responses and fail to capture CBT as a longitudinal, cross-session therapeutic process. To overcome this, we propose the DMT-CBT framework, which formalizes CBT for the first time as a partially observable therapeutic state evolution problem. DMT-CBT maintains structured cross-session states, integrates multimodal behavioral representations—including image-anchored cues—and incorporates a tool-augmented intervention mechanism. We also introduce DMTCorpus, the first synthetic multi-session CBT dataset. Experimental results demonstrate that DMT-CBT significantly outperforms post-hoc state extraction baselines in counseling fidelity, therapeutic alliance quality, and positivity of emotional trajectories, while more accurately preserving consistency in therapeutic state progression.

Abstract

Large language models (LLMs) have shown growing potential for Cognitive Behavioral Therapy (CBT) counseling. However, most existing approaches still formulate counseling as a local response generation problem, focusing on empathetic replies within short, text-only, or single-session interactions. We argue that this formulation fundamentally mismatches the nature of real psychotherapy. In clinical CBT, therapy is a longitudinal process in which therapists continuously infer, update, and intervene on evolving therapeutic states across sessions. Realistic CBT further involves multimodal inference and delayed cross-session intervention effects, requiring models to capture longitudinal therapeutic state evolution under partial observability. We propose DMT-CBT, a framework for Dynamic Modeling of evolving Therapeutic states in CBT counseling. DMT-CBT maintains structured therapeutic states across sessions while incorporating multimodal behavioral grounding and tool-augmented intervention to support adaptive therapeutic reasoning. Based on this framework, we construct DMTCorpus, a synthetic multi-session multimodal CBT counseling dataset featuring evolving therapeutic states, image-grounded client behaviors, and cross-session intervention continuity. Experimental results show that DMT-CBT improves counseling fidelity and therapeutic alliance, produces more favorable longitudinal affective trajectories, and preserves therapeutic states more faithfully than post-hoc extraction approaches.

Problem

• Cognitive Behavioral Therapy
• longitudinal therapeutic state
• multimodal inference
• cross-session intervention
• partial observability

Innovation

• longitudinal therapeutic modeling
• multimodal behavioral grounding
• tool-augmented intervention
• therapeutic state evolution
• CBT counseling