Advancing Conversational Psychotherapy: Integrating Privacy, Dual-Memory, and Domain Expertise with Large Language Models

TL;DR

SoulSpeak tackles accessibility and privacy barriers in psychotherapy by presenting an LLM-based chatbot with a dual-memory architecture and a privacy module. It uses a Retrieval Augmented Generation pipeline plus a therapist-domain knowledge base (Counsel Chat) to align responses with psychotherapeutic methods, and introduces the Conversational Psychotherapy Preference Model (CPPM) to simulate user preferences with high accuracy. Experimental results show SoulSpeak can outperform the lowest-preference human responses and approach the quality of high-preference therapists, while long-term memory improves relevance, though privacy, data availability, and clinical validation remain important considerations. The work highlights a practical path toward private, personalized, scalable mental-health support and provides evaluation tools that can be applied to other psychotherapy-focused LLM systems.

Abstract

Mental health has increasingly become a global issue that reveals the limitations of traditional conversational psychotherapy, constrained by location, time, expense, and privacy concerns. In response to these challenges, we introduce SoulSpeak, a Large Language Model (LLM)-enabled chatbot designed to democratize access to psychotherapy. SoulSpeak improves upon the capabilities of standard LLM-enabled chatbots by incorporating a novel dual-memory component that combines short-term and long-term context via Retrieval Augmented Generation (RAG) to offer personalized responses while ensuring the preservation of user privacy and intimacy through a dedicated privacy module. In addition, it leverages a counseling chat dataset of therapist-client interactions and various prompting techniques to align the generated responses with psychotherapeutic methods. We introduce two fine-tuned BERT models to evaluate the system against existing LLMs and human therapists: the Conversational Psychotherapy Preference Model (CPPM) to simulate human preference among responses and another to assess response relevance to user input. CPPM is useful for training and evaluating psychotherapy-focused language models independent from SoulSpeak, helping with the constrained resources available for psychotherapy. Furthermore, the effectiveness of the dual-memory component and the robustness of the privacy module are also examined. Our findings highlight the potential and challenge of enhancing mental health care by offering an alternative that combines the expertise of traditional therapy with the advantages of LLMs, providing a promising way to address the accessibility and personalization gap in current mental health services.

Figures (17)

• Figure 1: The SoulSpeak system architecture and user workflow.
• Figure 2: Overview of the Conversational Psychotherapy Preference Model (CPPM) to simulate service users' preference over responses. Note that the two responses as input to the model are assumed to be for the same question or query.
• Figure 3: Simulation of user preference on generated responses and therapists' responses. Human-worst/human-best represents the set of responses with the lowest/highest preference score for each question in the test set.
• Figure 4: Statistical comparison between the therapist's and SoulSpeak generated responses.
• Figure 5: Example of one entity and its summary in SoulSpeak's entity store. The entity name is anonymized by the privacy module. The entity summary is generated by an LLM API call given the context of the previous therapy session.