Towards Enhancing Health Coaching Dialogue in Low-Resource Settings

TL;DR

This paper addresses the high cost of health coaching by proposing a modular dialogue system that mimics goal-setting and progress follow-up while addressing patient emotions with empathy. It combines a simplified NLU for online belief tracking ($B_t$) and a stage-guided NLG ($R_t$) with an auxiliary, mechanism-conditioned empathetic generator that leverages external empathetic data. Empirical results show improved slot-filling and goal-tracking accuracy, and superior dialogue generation and empathy scores compared with baselines, validated by expert human evaluation. The work advances automated, accessible health coaching for low-resource settings and lays groundwork for end-to-end empathetic dialogue systems in clinical behavior-change applications.

Abstract

Health coaching helps patients identify and accomplish lifestyle-related goals, effectively improving the control of chronic diseases and mitigating mental health conditions. However, health coaching is cost-prohibitive due to its highly personalized and labor-intensive nature. In this paper, we propose to build a dialogue system that converses with the patients, helps them create and accomplish specific goals, and can address their emotions with empathy. However, building such a system is challenging since real-world health coaching datasets are limited and empathy is subtle. Thus, we propose a modularized health coaching dialogue system with simplified NLU and NLG frameworks combined with mechanism-conditioned empathetic response generation. Through automatic and human evaluation, we show that our system generates more empathetic, fluent, and coherent responses and outperforms the state-of-the-art in NLU tasks while requiring less annotation. We view our approach as a key step towards building automated and more accessible health coaching systems.

Figures (3)

• Figure 1: A health coaching dialogue scenario in our dialogue system. It starts with a goal-setting stage where the coach discusses a realistic goal with the patient. After the goal is settled, the coach follows up on the patient's progress and maintains patient engagement. Understanding the patient's emotion cues and responding empathetically is also crucial in such a scenario; the struck-out response, generated by a naïve sequence-to-sequence model, is inappropriate without the capability of understanding and modeling empathy.
• Figure 2: The framework of our health coaching dialogue system. The NLU module consisting of the slot-filling and carryover model reads the dialogue and infers belief state $B_t$. The NLG hc module takes as input the stage $S_{t-1}$, belief state $B_t$, and context $C_t$ to generate response $R_t$. The NLG emp handles the cases where empathy are required and outputs empathetic response $\tilde{R}_t$ and emotion signal $E_t$.
• Figure 3: Model architecture of generating $R_t$ in NLG hc. The context $C_t$, belief tokens $B_t$, and the predicted stage $S_t$ are concatenated as a single input sequence, training with a T5 encoder-decoder model.