ERD: A Framework for Improving LLM Reasoning for Cognitive Distortion Classification

TL;DR

This paper introduces ERD, a framework that enhances large language models' ability to classify cognitive distortions in psychotherapy by extracting relevant parts and debating reasoning steps among multiple agents, leading to improved accuracy and debiasing.

Contribution

ERD is a novel framework that combines extraction and multi-agent debate modules to improve LLM-based cognitive distortion classification performance.

Findings

ERD improves multi-class F1 score on a public dataset.

ERD enhances binary specificity score, reducing false positives.

Multi-agent debate helps debias the baseline model.

Abstract

Improving the accessibility of psychotherapy with the aid of Large Language Models (LLMs) is garnering a significant attention in recent years. Recognizing cognitive distortions from the interviewee's utterances can be an essential part of psychotherapy, especially for cognitive behavioral therapy. In this paper, we propose ERD, which improves LLM-based cognitive distortion classification performance with the aid of additional modules of (1) extracting the parts related to cognitive distortion, and (2) debating the reasoning steps by multiple agents. Our experimental results on a public dataset show that ERD improves the multi-class F1 score as well as binary specificity score. Regarding the latter score, it turns out that our method is effective in debiasing the baseline method which has high false positive rate, especially when the summary of multi-agent debate is provided to LLMs.