MentraSuite: Post-Training Large Language Models for Mental Health Reasoning and Assessment

2025-12-10

AI Summary

Large language models (LLMs) in mental health face critical challenges including incoherent reasoning, hallucination susceptibility, and poor clinical alignment. To address these, we propose Mindora—a novel framework featuring (1) MentraBench, the first five-dimensional benchmark for reasoning quality in mental health, evaluating concision, coherence, hallucination resistance, task comprehension, and internal consistency; and (2) a hybrid supervised fine-tuning–reinforcement learning (SFT-RL) post-training method incorporating inconsistency-detection rewards and structured reasoning trajectory generation to enable stepwise, verifiable, and clinically grounded inference. Experiments demonstrate that Mindora achieves top-ranked performance on MentraBench and significantly outperforms 20 baseline models on complex clinical tasks—including clinical abstraction, diagnostic validation, and intervention planning—establishing a trustworthy, clinically informed reasoning paradigm for mental health AI.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Mental health disorders affect hundreds of millions globally, and the Web now serves as a primary medium for accessing support, information, and assessment. Large language models (LLMs) offer scalable and accessible assistance, yet their deployment in mental-health settings remains risky when their reasoning is incomplete, inconsistent, or ungrounded. Existing psychological LLMs emphasize emotional understanding or knowledge recall but overlook the step-wise, clinically aligned reasoning required for appraisal, diagnosis, intervention planning, abstraction, and verification. To address these issues, we introduce MentraSuite, a unified framework for advancing reliable mental-health reasoning. We propose MentraBench, a comprehensive benchmark spanning five core reasoning aspects, six tasks, and 13 datasets, evaluating both task performance and reasoning quality across five dimensions: conciseness, coherence, hallucination avoidance, task understanding, and internal consistency. We further present Mindora, a post-trained model optimized through a hybrid SFT-RL framework with an inconsistency-detection reward to enforce faithful and coherent reasoning. To support training, we construct high-quality trajectories using a novel reasoning trajectory generation strategy, that strategically filters difficult samples and applies a structured, consistency-oriented rewriting process to produce concise, readable, and well-balanced trajectories. Across 20 evaluated LLMs, Mindora achieves the highest average performance on MentraBench and shows remarkable performances in reasoning reliability, demonstrating its effectiveness for complex mental-health scenarios.

Problem

• Develops a framework for reliable mental health reasoning in LLMs
• Addresses incomplete and inconsistent reasoning in mental health applications
• Enhances step-wise clinical reasoning for assessment and diagnosis

Innovation

• Post-trained LLM with hybrid SFT-RL framework
• Inconsistency-detection reward for faithful reasoning
• Novel trajectory generation with filtering and rewriting