Enhancing Psychological Counseling with Large Language Model: A Multifaceted Decision-Support System for Non-Professionals (Preprint)

Abstract

BACKGROUND In the contemporary landscape of social media, a large number of users express negative emotions, including strong suicidal intentions. The situation underscores the critical need for trained psychological counselors to enact effective mental interventions. However, the training of such professionals is often time-consuming, leading to a pressing need for mobilizing non-professionals or volunteers. OBJECTIVE This paper aims to introduce a novel model built on the foundation of large language models to fully assist non-professionals in providing psychological interventions in online user discourses. METHODS We've developed a specialized counseling support system aimed at enhancing communication between counselors and individuals. Built on a robust Large Language Model (LLM) framework, the system prioritizes user privacy through well-defined policies. It employs meticulously crafted prompts to steer the LLM in assisting counselors with problem analysis and in recommending effective communication strategies. To further safeguard against potential harm, all final feedback undergoes rigorous filtering to eliminate the possibility of generating detrimental comments. RESULTS We conducted a comprehensive evaluation involving ten professional psychological counselors with varying levels of expertise. The findings indicate that our system achieves more than 93% expert agreement on its accuracy in analyzing patients' issues. Moreover, it offers satisfactory treatment strategies, as demonstrated by a 78% expert approval rate. CONCLUSIONS The research serves as a validation of the application of large language models in the field of psychology. It also lays the groundwork for a new paradigm in community-based mental health support, enabling the meaningful mobilization of non-professional counselors.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 14 citations exist upstream.