Response-act Guided Reinforced Dialogue Generation for Mental Health Counseling

TL;DR

This paper introduces READER, a transformer-based dialogue generation model guided by dialogue-acts and reinforcement learning, to improve mental health counseling conversations in virtual assistants.

Contribution

It presents a novel response-act guided reinforcement learning approach for dialogue generation, specifically tailored for mental health counseling applications.

Findings

READER outperforms baseline models on HOPE dataset

Achieves higher METEOR, ROUGE, and BERTScore metrics

Demonstrates effective integration of dialogue-acts and semantic richness

Abstract

Virtual Mental Health Assistants (VMHAs) have become a prevalent method for receiving mental health counseling in the digital healthcare space. An assistive counseling conversation commences with natural open-ended topics to familiarize the client with the environment and later converges into more fine-grained domain-specific topics. Unlike other conversational systems, which are categorized as open-domain or task-oriented systems, VMHAs possess a hybrid conversational flow. These counseling bots need to comprehend various aspects of the conversation, such as dialogue-acts, intents, etc., to engage the client in an effective conversation. Although the surge in digital health research highlights applications of many general-purpose response generation systems, they are barely suitable in the mental health domain -- the prime reason is the lack of understanding in mental health counseling. Moreover, in general, dialogue-act guided response generators are either limited to a template-based paradigm or lack appropriate semantics. To this end, we propose READER -- a REsponse-Act guided reinforced Dialogue genERation model for the mental health counseling conversations. READER is built on transformer to jointly predict a potential dialogue-act d(t+1) for the next utterance (aka response-act) and to generate an appropriate response u(t+1). Through the transformer-reinforcement-learning (TRL) with Proximal Policy Optimization (PPO), we guide the response generator to abide by d(t+1) and ensure the semantic richness of the responses via BERTScore in our reward computation. We evaluate READER on HOPE, a benchmark counseling conversation dataset and observe that it outperforms several baselines across several evaluation metrics -- METEOR, ROUGE, and BERTScore. We also furnish extensive qualitative and quantitative analyses on results, including error analysis, human evaluation, etc.