Psychological Assessments with Large Language Models: A Privacy-Focused and Cost-Effective Approach

TL;DR

This work investigates privacy-preserving, cost-effective psychological assessment using open-source LLMs that run locally on commodity hardware. It demonstrates a simple prompting strategy plus a grammar-based constraint to extract evidence and generate summaries supporting preassigned suicide-risk levels for Reddit comments, evaluated within the CLPsych 2024 task. OpenHermes performs particularly well for summaries and shows competitive results for highlights under a non-tuned, low-resource regime, highlighting the feasibility of privacy-conscious NLP in clinical psychology. The findings suggest that open-source, locally deployable models can deliver meaningful, ethically responsible assessments without relying on third-party APIs or extensive fine-tuning, with implications for scalable, privacy-preserving mental health analytics.

Abstract

This study explores the use of Large Language Models (LLMs) to analyze text comments from Reddit users, aiming to achieve two primary objectives: firstly, to pinpoint critical excerpts that support a predefined psychological assessment of suicidal risk; and secondly, to summarize the material to substantiate the preassigned suicidal risk level. The work is circumscribed to the use of "open-source" LLMs that can be run locally, thereby enhancing data privacy. Furthermore, it prioritizes models with low computational requirements, making it accessible to both individuals and institutions operating on limited computing budgets. The implemented strategy only relies on a carefully crafted prompt and a grammar to guide the LLM's text completion. Despite its simplicity, the evaluation metrics show outstanding results, making it a valuable privacy-focused and cost-effective approach. This work is part of the Computational Linguistics and Clinical Psychology (CLPsych) 2024 shared task.