Work

Building Psychotherapy Awareness in LLM: A Novel GenAI Multi-Agent Framework for Predicting Client’s Remission

Year: 2026

Type: article

Abstract: Motivational Interviewing (MI) is a widely recognised therapeutic approach for behaviour modification in clients suffering from addiction. An important step in evaluating the effectiveness of MI involves annotating the clients’ utterances and therapists’ behaviour associated with the utterances. Manually annotating an MI transcript is often time-consuming, whereas automating it with Large Language Models (LLMs) leads to a drop in quality. This paper advances the use of LLMs in clinical applications by presenting a novel multi-agent framework for automatically annotating MI transcripts and predicting clients’ remission. The framework consists of multiple LLM-agents operating in two stages by decomposing the complex task of predicting clients’ remission into simpler subtasks that individual agents can manage. In stage one, a set of reasoning agents generate reasons for annotating utterances by adhering to the Motivational Interviewing Treatment Integrity (MITI). These reasons are used by decision agents to annotate an utterance and create a MI-aware summary of their decision. The MI-aware summary captures the clients’ motivation and resilience to change as induced by the therapist. Once the transcript is annotated, stage two begins with another set of reasoning agents generating reasons for remit or non-remit prediction. A final decision agent then uses the reasoning to predict clients’ remission. The proposed framework outperforms the state of the art, achieving 89% annotation accuracy on the AnnoMI dataset. To evaluate the framework’s ability to predict remission, this paper introduces a novel Clinical MI dataset. The proposed framework achieves 85% accuracy in predicting remission without any training or fine-tuning.

Source: IEEE Access

DOI: 10.1109/access.2026.3727051

Authors: Dinesh Rivankar, Tanmay Tulsidas Verlekar, Abhijit Nadkarni

Institutions: Birla Institute of Technology and Science, Pilani - Goa Campus, Birla Institute of Technology and Science, Pilani, London School of Hygiene & Tropical Medicine

Language: English

Topic: Mental Health via Writing

Subfield: Social Psychology

Field: Psychology

Domain: Social Sciences

SDG: Good health and well-being

Open Access status: gold

Estimated APC paid: $2,075