Towards Efficient and Robust Linguistic Emotion Diagnosis for Mental Health via Multi-Agent Instruction Refinement

TL;DR

This work presents APOLO, a POMDP-based multi-agent framework for Automated Prompt Optimization in linguistic emotion diagnosis. By coordinating a Planner with a Socratic Teacher–Critic–Student loop and a Target evaluator, APOLO achieves risk- and cost-aware prompt refinement that improves multi-label emotion coverage and diagnostic reliability in medical text. Empirical results across six benchmarks and multiple backbones show state-of-the-art performance, data efficiency (even in 1-shot settings), and favorable inference-time scaling. The approach offers a scalable, interpretable pathway to trustworthy AI-assisted mental-health assessment in real-world clinical contexts.

Abstract

Linguistic expressions of emotions such as depression, anxiety, and trauma-related states are pervasive in clinical notes, counseling dialogues, and online mental health communities, and accurate recognition of these emotions is essential for clinical triage, risk assessment, and timely intervention. Although large language models (LLMs) have demonstrated strong generalization ability in emotion analysis tasks, their diagnostic reliability in high-stakes, context-intensive medical settings remains highly sensitive to prompt design. Moreover, existing methods face two key challenges: emotional comorbidity, in which multiple intertwined emotional states complicate prediction, and inefficient exploration of clinically relevant cues. To address these challenges, we propose APOLO (Automated Prompt Optimization for Linguistic Emotion Diagnosis), a framework that systematically explores a broader and finer-grained prompt space to improve diagnostic efficiency and robustness. APOLO formulates instruction refinement as a Partially Observable Markov Decision Process and adopts a multi-agent collaboration mechanism involving Planner, Teacher, Critic, Student, and Target roles. Within this closed-loop framework, the Planner defines an optimization trajectory, while the Teacher-Critic-Student agents iteratively refine prompts to enhance reasoning stability and effectiveness, and the Target agent determines whether to continue optimization based on performance evaluation. Experimental results show that APOLO consistently improves diagnostic accuracy and robustness across domain-specific and stratified benchmarks, demonstrating a scalable and generalizable paradigm for trustworthy LLM applications in mental healthcare.

Figures (6)

• Figure 1: Examples of disease-related emotion diagnosis under three prompting strategies: zero-shot, CoT, and APOLO.
• Figure 2: Comparison of APO strategies. Top: generation–search methods generate and locally refine candidate prompts, leading to limited coverage. Middle: meta-prompt approaches rely on fixed optimization templates with low adaptability. Bottom: our APOLO framework introduces multi-agent adaptive reasoning, enabling dynamic and collaborative prompt exploration. Right: performance comparison across three datasets based on m-F1 score.
• Figure 3: The overall architecture of the APOLO model, designed to support medical emotion diagnosis tasks. It consists of five LLM agents. The Planner agent that autonomously generates task-specific optimization trajectories, and a Teacher-Critic-Student Socratic dialogue mechanism that iteratively refines prompts, with the evaluation and iterative refinement process guided by feedback from the Target agent.
• Figure 4: Inference-time scaling law for APOLO and baseline methods, evaluated on datasets related to medical emotion diagnosis, illustrating the relationship between Micro F1-score and total token consumption during optimization.
• Figure 5: Convergence analysis of APOLO's optimization process on six emotion diagnosis datasets. The plots track the Micro F1-score (m-F1) over 10 iterations. APOLO's trajectory is compared against the iterative baseline OPRO and three static baselines (Original, CoT, APE).