SciGroveBeta
Medicine

Mental-R1: Aligning LLM Reasoning for Mental Health Assessment

Xin Wang, Boyan Gao, Yibo Yang, David A. Clifton

Featured July 6, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

By teaching large language models to think like humans, starting with exploratory uncertainty and moving to confident conclusions through structured steps, the paper helps them better assess mental health from text.

In depth
The paper introduces Cognitive Relative Policy Optimization (CRPO), a reinforcement learning framework that aligns large language model (LLM) reasoning with human cognitive processes for mental health assessment. It achieves this by integrating stage-dependent uncertainty modeling (Stage-wise Entropy Regularization) to mimic the human shift from exploration to confident decision-making, and by formalizing theory-grounded cognitive reasoning stages inspired by cognitive appraisal theory.

Key Takeaways

  • 1
    CRPO introduces Stage-wise Entropy Regularization to guide LLMs from exploratory uncertainty in early reasoning stages to confident decision-making in later stages, mirroring human cognition.
  • 2
    The framework formalizes cognitive reasoning stages (Stimulus, Primary Appraisal, Secondary Appraisal, Reaction, Mental State) based on cognitive appraisal theory, enabling more interpretable and nuanced mental health assessments.
  • 3
    Experiments on 8 mental health datasets demonstrate that CRPO-trained models achieve an average 10.4 percentage point improvement in weighted F1-score over the best reinforcement learning baselines, particularly excelling in reasoning-intensive cases.

Conceptual Flow

HIGH LEVEL
1
Methodology: Human-like Reasoning for LLMs

The new method teaches computer models to think step-by-step, just like people do when figuring things out, starting broad and then getting specific.

Input Text
Think Step-by-Step
Mental State Prediction
2
Results: Better Mental Health Assessment

This new way of thinking helps the computer models make much more accurate predictions about mental health, especially for tricky cases.

Old Prediction
New Prediction
Compare Accuracy
Much Better Results