Selected work

Representative Papers

PsychePass: Calibrating LLM Therapeutic Competence via Trajectory-Anchored Tournaments

Jan 28, 2026

Existing evaluation methods struggle to effectively assess the therapeutic capabilities of large language models, often compromised by process drift and criterion drift. To address this, this work proposes a framework integrating trajectory-anchored client simulation with a Swiss-system dynamic tournament, coupled with an Elo rating mechanism to transform interaction trajectories into stable, comparable reward signals for policy gradient reinforcement learning. This approach enables unified calibration and efficient optimization of models’ therapeutic skills. Experimental results demonstrate that the framework’s evaluations align closely with human expert judgments and significantly enhance model performance in psychotherapeutic tasks.

0 citationsRead paper

Unveiling the Landscape of Clinical Depression Assessment: From Behavioral Signatures to Psychiatric Reasoning

Aug 06, 2025

Existing depression assessment methods rely heavily on non-clinical data and employ complex models with limited clinical deployability. Method: This work proposes a clinically grounded, multimodal depression assessment framework tailored to real-world diagnostic settings. We introduce the C-MIND dataset—comprising synchronized audio, video, transcribed text, and functional near-infrared spectroscopy (fNIRS) neuroimaging signals—and design a clinical-knowledge-guided large language model (LLM) inference mechanism that jointly integrates behavioral analysis and psychiatric diagnostic reasoning. Furthermore, we systematically quantify the contribution of each modality to the diagnostic task. Results: Evaluated on authentic clinical data, our approach achieves a 10% improvement in Macro-F1 score over prior methods, while significantly enhancing model interpretability and practical deployability. To our knowledge, this is the first automated depression assessment framework grounded in real clinical workflows, balancing diagnostic reliability with clinical utility.

0 citationsRead paper

Reframe Your Life Story: Interactive Narrative Therapist and Innovative Moment Assessment with Large Language Models

Jul 27, 2025

Existing large language models (LLMs) lack domain-specific therapeutic simulation capabilities and fail to track longitudinal healing progress in mental health support. Method: We propose the first interactive LLM system designed specifically for narrative therapy, integrating treatment-stage planning, reflective-level scaffolding, context-aware response generation, and narrative transformation detection to form a closed-loop intervention framework. We further introduce a novel “Moment of Innovation” quantification model to dynamically monitor narrative reconstruction and deliver personalized feedback. Contribution/Results: Evaluated on 260 simulated dialogues and 230 human participants, the system significantly enhances dialogue depth and therapeutic quality (p < 0.01), generating more empowering and socially supportive responses. It overcomes key limitations of general-purpose LLMs—namely insufficient therapeutic depth, poor temporal continuity, and low clinical adaptability—thereby advancing AI’s role in evidence-informed, narrative-based mental health interventions.

0 citationsRead paper

{Psi}-Arena: Interactive Assessment and Optimization of LLM-based Psychological Counselors with Tripartite Feedback

May 06, 2025

Existing evaluation methods for LLM-based counseling agents suffer from static design, single-perspective assessment, and non-actionable feedback. Method: We propose an interactive evaluation and optimization framework featuring (1) multi-stage NPC dialogues grounded in psychological profiling to simulate authentic counseling scenarios; (2) a novel tripartite collaborative evaluation mechanism involving users, AI, and human experts; and (3) a diagnosis-driven, closed-loop reflective optimization paradigm integrating reflection-based RLHF and structured feedback generation. Contribution/Results: Evaluated on eight mainstream LLMs, our framework reveals significant inter-model capability disparities; reflective optimization improves counseling quality by up to 141%. We release the first reproducible, extensible benchmark platform for mental health LLMs—advancing LLM-powered counseling toward safety, trustworthiness, and human-centered alignment.

0 citationsRead paper

MAGI: Multi-Agent Guided Interview for Psychiatric Assessment

Apr 25, 2025

Current large language model (LLM)-based approaches struggle to align with standardized psychiatric diagnostic protocols, limiting the clinical utility of automated mental health interviews for improving access to care. To address this, we propose the first multi-agent system for automating the Mini-International Neuropsychiatric Interview (MINI), uniquely decomposing its clinical protocol into four synergistic agent roles: clinical logic navigation, adaptive question generation, response classification, and diagnostic reasoning. We introduce Psychometric Chain-of-Thought (PsyCoT), a novel mechanism enabling interpretable mapping from symptom representations to DSM/ICD diagnostic criteria. The system integrates decision-tree–guided navigation, empathetic dialogue generation, and structured response validation. Evaluated on 1,002 real-world participants, it performs assessments for major depressive disorder, generalized anxiety disorder, social anxiety disorder, and suicide risk. Experiments demonstrate significant improvements in clinical consistency (+23.6%), conversational adaptability (+31.4%), and diagnostic interpretability.

0 citationsRead paper

Recent publications

Latest Papers

PsychePass: Calibrating LLM Therapeutic Competence via Trajectory-Anchored Tournaments

Jan 28, 2026

Existing evaluation methods struggle to effectively assess the therapeutic capabilities of large language models, often compromised by process drift and criterion drift. To address this, this work proposes a framework integrating trajectory-anchored client simulation with a Swiss-system dynamic tournament, coupled with an Elo rating mechanism to transform interaction trajectories into stable, comparable reward signals for policy gradient reinforcement learning. This approach enables unified calibration and efficient optimization of models’ therapeutic skills. Experimental results demonstrate that the framework’s evaluations align closely with human expert judgments and significantly enhance model performance in psychotherapeutic tasks.

0 citationsRead paper

Unveiling the Landscape of Clinical Depression Assessment: From Behavioral Signatures to Psychiatric Reasoning

Aug 06, 2025

Existing depression assessment methods rely heavily on non-clinical data and employ complex models with limited clinical deployability. Method: This work proposes a clinically grounded, multimodal depression assessment framework tailored to real-world diagnostic settings. We introduce the C-MIND dataset—comprising synchronized audio, video, transcribed text, and functional near-infrared spectroscopy (fNIRS) neuroimaging signals—and design a clinical-knowledge-guided large language model (LLM) inference mechanism that jointly integrates behavioral analysis and psychiatric diagnostic reasoning. Furthermore, we systematically quantify the contribution of each modality to the diagnostic task. Results: Evaluated on authentic clinical data, our approach achieves a 10% improvement in Macro-F1 score over prior methods, while significantly enhancing model interpretability and practical deployability. To our knowledge, this is the first automated depression assessment framework grounded in real clinical workflows, balancing diagnostic reliability with clinical utility.

0 citationsRead paper

Reframe Your Life Story: Interactive Narrative Therapist and Innovative Moment Assessment with Large Language Models

Jul 27, 2025

Existing large language models (LLMs) lack domain-specific therapeutic simulation capabilities and fail to track longitudinal healing progress in mental health support. Method: We propose the first interactive LLM system designed specifically for narrative therapy, integrating treatment-stage planning, reflective-level scaffolding, context-aware response generation, and narrative transformation detection to form a closed-loop intervention framework. We further introduce a novel “Moment of Innovation” quantification model to dynamically monitor narrative reconstruction and deliver personalized feedback. Contribution/Results: Evaluated on 260 simulated dialogues and 230 human participants, the system significantly enhances dialogue depth and therapeutic quality (p < 0.01), generating more empowering and socially supportive responses. It overcomes key limitations of general-purpose LLMs—namely insufficient therapeutic depth, poor temporal continuity, and low clinical adaptability—thereby advancing AI’s role in evidence-informed, narrative-based mental health interventions.

0 citationsRead paper

{Psi}-Arena: Interactive Assessment and Optimization of LLM-based Psychological Counselors with Tripartite Feedback

May 06, 2025

Existing evaluation methods for LLM-based counseling agents suffer from static design, single-perspective assessment, and non-actionable feedback. Method: We propose an interactive evaluation and optimization framework featuring (1) multi-stage NPC dialogues grounded in psychological profiling to simulate authentic counseling scenarios; (2) a novel tripartite collaborative evaluation mechanism involving users, AI, and human experts; and (3) a diagnosis-driven, closed-loop reflective optimization paradigm integrating reflection-based RLHF and structured feedback generation. Contribution/Results: Evaluated on eight mainstream LLMs, our framework reveals significant inter-model capability disparities; reflective optimization improves counseling quality by up to 141%. We release the first reproducible, extensible benchmark platform for mental health LLMs—advancing LLM-powered counseling toward safety, trustworthiness, and human-centered alignment.

0 citationsRead paper

MAGI: Multi-Agent Guided Interview for Psychiatric Assessment

Apr 25, 2025

Current large language model (LLM)-based approaches struggle to align with standardized psychiatric diagnostic protocols, limiting the clinical utility of automated mental health interviews for improving access to care. To address this, we propose the first multi-agent system for automating the Mini-International Neuropsychiatric Interview (MINI), uniquely decomposing its clinical protocol into four synergistic agent roles: clinical logic navigation, adaptive question generation, response classification, and diagnostic reasoning. We introduce Psychometric Chain-of-Thought (PsyCoT), a novel mechanism enabling interpretable mapping from symptom representations to DSM/ICD diagnostic criteria. The system integrates decision-tree–guided navigation, empathetic dialogue generation, and structured response validation. Evaluated on 1,002 real-world participants, it performs assessments for major depressive disorder, generalized anxiety disorder, social anxiety disorder, and suicide risk. Experiments demonstrate significant improvements in clinical consistency (+23.6%), conversational adaptability (+31.4%), and diagnostic interpretability.

0 citationsRead paper