Sentiment-guided Commonsense-aware Response Generation for Mental Health Counseling

TL;DR

The paper introduces EmpRes, a sentiment-guided commonsense-aware framework for therapist response generation in Virtual Mental Health Assistants. It combines a Speaker-Context Graph and a Sentiment-Guided Commonsense Relation Graph with a GPT-2 decoder equipped with Knowledge Aware Attention to produce responses that shape client sentiment toward positivity. Evaluations on the HOPE dataset show EmpRes outperforms twelve baselines across standard text-generation metrics and gains strong human-validation support, even nearing or surpassing gold standards in some cases. A real-world deployment and user study demonstrate high perceived effectiveness and willingness to continue using the system, while acknowledging limitations such as occasional hallucinations and data requirements for broader scalability.

Abstract

The crisis of mental health issues is escalating. Effective counseling serves as a critical lifeline for individuals suffering from conditions like PTSD, stress, etc. Therapists forge a crucial therapeutic bond with clients, steering them towards positivity. Unfortunately, the massive shortage of professionals, high costs, and mental health stigma pose significant barriers to consulting therapists. As a substitute, Virtual Mental Health Assistants (VMHAs) have emerged in the digital healthcare space. However, most existing VMHAs lack the commonsense to understand the nuanced sentiments of clients to generate effective responses. To this end, we propose EmpRes, a novel sentiment-guided mechanism incorporating commonsense awareness for generating responses. By leveraging foundation models and harnessing commonsense knowledge, EmpRes aims to generate responses that effectively shape the client's sentiment towards positivity. We evaluate the performance of EmpRes on HOPE, a benchmark counseling dataset, and observe a remarkable performance improvement compared to the existing baselines across a suite of qualitative and quantitative metrics. Moreover, our extensive empirical analysis and human evaluation show that the generation ability of EmpRes is well-suited and, in some cases, surpasses the gold standard. Further, we deploy EmpRes as a chat interface for users seeking mental health support. We address the deployed system's effectiveness through an exhaustive user study with a significant positive response. Our findings show that 91% of users find the system effective, 80% express satisfaction, and over 85.45% convey a willingness to continue using the interface and recommend it to others, demonstrating the practical applicability of EmpRes in addressing the pressing challenges of mental health support, emphasizing user feedback, and ethical considerations in a real-world context.

Figures (9)

• Figure 1: A sample counseling conversation in which the client expresses negative sentiment at last; the job of a VMHA is to generate a response to positively influence the client and contains rich commonsense knowledge.
• Figure 2: A schematic diagram of pseudo labeling for sentiment labels for each utterance in the HOPE dataset. We augment top-5 additional attributes using COMET to assist state-of-the-art sentiment classifier in predicting sentiment label.
• Figure 3: A schematic diagram of EmpRes. Knowledge Extractor exploits conditional commonsense relations $(r_i)$ to construct Sentiment-Guided Commonsense Relation Graph (SGCR-Graph). A graph attention layer (GAT-SGCR), on top of SGCR-Graph, formulates sentiment-guided commonsense-aware representations. A similar layer on Speaker-Context Graph (SC-Graph), called GAT-SC, is responsible for formulating dialogue representations. Fused graph representations act as key and value for GPT-2's attention block, responsible for therapist response generation.
• Figure 4: The figure illustrates (A) the prototype deployment of the chat interface and (B) findings of the user study conducted to evaluate the real-world effectiveness of the deployed EmpRes. (C) The user study comprises user feedback of conversation with the deployed system across five major criteria -- (i) perceived effectiveness, (ii) user satisfaction, (iii) continued usage, (iv) likelihood to recommend, and (v) observed hallucination.
• Figure :