Enhancing LLM-Based Just-in-Time Support with Theoretical Guidance from Dialectical Behavior Therapy

Abstract

Just-in-time adaptive interventions promise support at the moment distress occurs, but the content of that support is often generated without reference to an established therapeutic framework. We examine how Dialectical Behavior Therapy (DBT) can provide theoretical guidance for large language model responses to contextual enquiries from college students. Grounding generation in DBT skills and principles offers a route to responses that are clinically coherent and appropriate to the person’s state, rather than generically supportive. We present our approach and early observations, and discuss implications for deploying theory-guided conversational support in digital mental health systems.

Manas Bedmutha

Ph.D. Candidate

Aaron Broukhim

Ph.D. Candidate

Isaac Bouchard

Ph.D. Student
(Co-Advised with Elizabeth Eikey, Public Health)

Nadir Weibel

Professor of Computer Science and Engineering