Research map / Large language models / Agents, code and math
MentalGuide: Towards Multi-Turn, State-Aware and Strategy-Driven Conversations for Mental Health Support
Jinwei He, Feng Lu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add expert state-strategy prior probabilities (ablation) | MentalGuide (Deepseek-V3) · Strategy correctness | accuracy (%): 38.68 -> 52.45 | Improved |
| add expert state-strategy prior probabilities (ablation) | MentalGuide (Qwen-3-Plus) · Strategy correctness | accuracy (%): 41.15 -> 48.25 | Improved |
| add expert prior knowledge (ablation) | MentalGuide (Deepseek-V3) · Strategic execution (multi-turn) | score 0-3: 2.35 -> 2.52 | No clear effect |
| add expert prior knowledge (ablation) | MentalGuide (Deepseek-V3) · Strategy distribution deviation per round | Mr: 8.46 -> 6.02 | Improved |
| add expert prior knowledge (ablation) | MentalGuide (Deepseek-V3) · Strategy distribution deviation per client state | Ms: 11.62 -> 5.11 | Improved |
| use LLM-ranking weights 300/200/100 instead of 72/64/36 (ablation) | MentalGuide · Strategy distribution deviation per round | Mr: 6.02 -> 8.29 | Hurt |
| use LLM-ranking weights 30/20/10 instead of 72/64/36 (ablation) | MentalGuide · Strategy distribution deviation per round | Mr: 6.02 -> 9.23 | Hurt |
| select k=1 final strategy instead of 2 (ablation) | MentalGuide · Strategy distribution deviation per round | Mr: 6.02 -> 10.91 | Hurt |
| select k=3 final strategies instead of 2 (ablation) | MentalGuide · Strategy distribution deviation per round | Mr: 6.02 -> 7.45 | Hurt |
| MentalGuide vs RolePlay on same LLM | RolePlay (Deepseek-V3) · Strategic execution (multi-turn) | score 0-3: 1.55 -> 2.52 | Improved |
| MentalGuide vs RolePlay on single-turn emotional empathy | RolePlay (Qwen-3-Plus) · Single-turn emotional empathy | score 0-3: 2.25 -> 2.58 | Improved |
| MentalGuide vs RolePlay and SoulChat in human expert ranking | MentalGuide · Human expert ranking (10 scenarios, 10 psychologists) | share of Rank 1: Rank 1 in 58% of cases; Rank 3 in 10% | Improved |
Closest papers by meaning
- Position: Beyond Assistance – Reimagining LLMs as Ethical and Adaptive Co-Creators in Mental Health Care (ICML 2025 · similarity 0.86)
- MindGuard: Guardrail Classifiers for Multi-Turn Mental Health Support (NeurIPS 2026 · similarity 0.86)
- TherapyGym: Evaluating and Aligning Clinical Fidelity and Safety in Therapy Chatbots (NeurIPS 2026 · similarity 0.85)
- AgentMental: An Interactive Multi-Agent Framework for Explainable and Adaptive Mental Health Assessment (AAAI 2026 · similarity 0.85)
- STAMPsy: Towards SpatioTemporal-Aware Mixed-Type Dialogues for Psychological Counseling (AAAI 2025 · similarity 0.84)
- An LLM-based Simulation Framework for Embodied Conversational Agents in Psychological Counseling (AAAI 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
