TheraMind: A Strategic and Adaptive Agent for Longitudinal Psychological Counseling

TL;DR

TheraMind addresses the gap between single-session LLM counseling and real-world psychotherapy by introducing a dual-loop architecture that separates intra-session dialogue management from cross-session therapeutic planning. The Intra-Session Loop perceptually interprets emotion, retrieves memories, and generates clinically grounded responses, while the Cross-Session Loop evaluates efficacy after each session and adaptively selects the next therapy. Key contributions include memory-augmented perception, dynamic response strategies, post-session efficacy evaluation, and adaptive therapy selection within a high-fidelity CPsyCounR-based simulation. In comprehensive simulations, TheraMind outperforms baselines on multi-session metrics such as Coherence, Empathy, and Therapeutic Attunement, demonstrating superior longitudinal therapeutic behavior and setting a new benchmark for automated psychological counseling systems.

Abstract

Large language models (LLMs) in psychological counseling have attracted increasing attention. However, existing approaches often lack emotional understanding, adaptive strategies, and the use of therapeutic methods across multiple sessions with long-term memory, leaving them far from real clinical practice. To address these critical gaps, we introduce TheraMind, a strategic and adaptive agent for longitudinal psychological counseling. The cornerstone of TheraMind is a novel dual-loop architecture that decouples the complex counseling process into an Intra-Session Loop for tactical dialogue management and a Cross-Session Loop for strategic therapeutic planning. The Intra-Session Loop perceives the patient's emotional state to dynamically select response strategies while leveraging cross-session memory to ensure continuity. Crucially, the Cross-Session Loop empowers the agent with long-term adaptability by evaluating the efficacy of the applied therapy after each session and adjusting the method for subsequent interactions. We validate our approach in a high-fidelity simulation environment grounded in real clinical cases. Extensive evaluations show that TheraMind outperforms other methods, especially on multi-session metrics like Coherence, Flexibility, and Therapeutic Attunement, validating the effectiveness of its dual-loop design in emulating strategic, adaptive, and longitudinal therapeutic behavior. The code is publicly available at https://0mwwm0.github.io/TheraMind/.

Figures (30)

• Figure 1: Illustration of Our TheraMind Framework.
• Figure 2: The TheraMind framework operates on a novel dual-loop paradigm. The Intra-Session Loop (top) executes turn-by-turn dialogue management, involving patient state perception, memory retrieval, and clinically-grounded response generation. The Cross-Session Loop (bottom) performs macro-level strategic planning by evaluating therapeutic efficacy after each session and adaptively selecting the therapeutic method for the next, enabling true longitudinal and personalized counseling.
• Figure 3: Ablation Study. Pairwise comparison between the full TheraMind model and its ablated variants.
• Figure 4: Performance comparison across ten distinct categories of psychological issues, evaluated in both single-session (left) and multi-session (right) contexts. The axes correspond to the categories detailed in Table \ref{['tab:patient_categories']}.
• Figure 5: (Left) Human agreement analysis of TheraMind's internal decision-making modules. (Middle) Pairwise human preference evaluation. TheraMind was compared against three top-performing models on a simplified multi-session evaluation standard. (Right) Distribution of patient emotions recognized by TheraMind during counseling sessions.