Navigating through the hidden embedding space: steering LLMs to improve mental health assessment

2025-10-18

AI Summary

This study addresses the limited capability of small-scale large language models (LLMs) in two critical mental health assessment tasks: detecting depression-related posts (relevance prediction) and automatically completing standardized psychological screening questionnaires (questionnaire completion). To overcome these limitations, we propose a lightweight, low-computation implicit-space steering method that operates without parameter fine-tuning. Instead, it learns linear transformations and task-specific directional steering vectors within the activation space of selected model layers, enabling interpretable and controllable intervention on internal representations. Evaluated on real-world Reddit user data, our approach significantly improves performance on both tasks. Results demonstrate its effectiveness in unlocking the domain-specific potential of resource-constrained LLMs, offering a novel, cost-efficient paradigm for adapting LLMs to clinical auxiliary assessment scenarios.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

The rapid evolution of Large Language Models (LLMs) is transforming AI, opening new opportunities in sensitive and high-impact areas such as Mental Health (MH). Yet, despite these advancements, recent evidence reveals that smaller-scale models still struggle to deliver optimal performance in domain-specific applications. In this study, we present a cost-efficient yet powerful approach to improve MH assessment capabilities of an LLM, without relying on any computationally intensive techniques. Our lightweight method consists of a linear transformation applied to a specific layer's activations, leveraging steering vectors to guide the model's output. Remarkably, this intervention enables the model to achieve improved results across two distinct tasks: (1) identifying whether a Reddit post is useful for detecting the presence or absence of depressive symptoms (relevance prediction task), and (2) completing a standardized psychological screening questionnaire for depression based on users' Reddit post history (questionnaire completion task). Results highlight the untapped potential of steering mechanisms as computationally efficient tools for LLMs' MH domain adaptation.

Problem

• Improving mental health assessment with LLMs
• Steering model outputs via lightweight transformations
• Enhancing depression detection from social media posts

Innovation

• Linear transformation applied to layer activations
• Steering vectors guide model's output
• Lightweight method improves mental health assessment