Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SafeTalkCoach: Diversity-Driven Multi-Agent Simulation for Parent-Teen Health Conversations

Benyamin Tabarsi, Wenbo Li, Tahreem Yasir, Aryan Santhosh Kumar, Laura Widman, Dongkuan Xu, Tiffany Barnes

TL;DR

SafeTalkCoach introduces a diversity-driven, three-agent framework for simulating parent–teen conversations about sexual health to overcome data scarcity and realism gaps. By integrating grounded guidelines, adaptive engagement control, and a hierarchical diversification pipeline (data, architecture, prompting), it yields a large, realistic, and guideline-adherent dataset of 1,495 dialogues across three topics. Automated and human evaluations indicate improvements in diversity, realism, and communication quality relative to baselines, while ablation studies confirm the value of each component. The work offers practical resources for AI research and health communication practice, with planned expansions to more topics and broader expert validation, alongside careful consideration of ethics and privacy.

Abstract

The importance of effective parent-child communication about sexual health is widely acknowledged, but real-world data on these conversations is scarce and challenging to collect, due to their private and sensitive nature. Although LLMs have been widely adopted in dialogue generation, they may deviate from best practices and frequently lack realism and diversity. We introduce SafeTalkCoach, a diversity-driven multi-agent dialogue generation framework that simulates parent-child conversations about sexual health, and present an accompanying dataset. SafeTalkCoach integrates crowd-sourced and synthesized scenarios, established sexual health guidelines, evidence-based personas, adaptive control modules, and hierarchical diversification. Through evaluations, we demonstrate that SafeTalkCoach generates diverse conversations while maintaining realism, communication quality, and controllability in practice. Our goal is that the SafeTalkCoach framework and the dataset support both AI research and health communications practices.

SafeTalkCoach: Diversity-Driven Multi-Agent Simulation for Parent-Teen Health Conversations

TL;DR

SafeTalkCoach introduces a diversity-driven, three-agent framework for simulating parent–teen conversations about sexual health to overcome data scarcity and realism gaps. By integrating grounded guidelines, adaptive engagement control, and a hierarchical diversification pipeline (data, architecture, prompting), it yields a large, realistic, and guideline-adherent dataset of 1,495 dialogues across three topics. Automated and human evaluations indicate improvements in diversity, realism, and communication quality relative to baselines, while ablation studies confirm the value of each component. The work offers practical resources for AI research and health communication practice, with planned expansions to more topics and broader expert validation, alongside careful consideration of ethics and privacy.

Abstract

The importance of effective parent-child communication about sexual health is widely acknowledged, but real-world data on these conversations is scarce and challenging to collect, due to their private and sensitive nature. Although LLMs have been widely adopted in dialogue generation, they may deviate from best practices and frequently lack realism and diversity. We introduce SafeTalkCoach, a diversity-driven multi-agent dialogue generation framework that simulates parent-child conversations about sexual health, and present an accompanying dataset. SafeTalkCoach integrates crowd-sourced and synthesized scenarios, established sexual health guidelines, evidence-based personas, adaptive control modules, and hierarchical diversification. Through evaluations, we demonstrate that SafeTalkCoach generates diverse conversations while maintaining realism, communication quality, and controllability in practice. Our goal is that the SafeTalkCoach framework and the dataset support both AI research and health communications practices.
Paper Structure (39 sections, 2 equations, 4 figures, 22 tables)

This paper contains 39 sections, 2 equations, 4 figures, 22 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Parent-Teen Health Communication and Data Challenges
  4. LLM-Based Dialogue Generation Frameworks
  5. Diverse Dialogue Generation
  6. The Proposed Method: SafeTalkCoach
  7. Multi-Agent Dialogue Generation Pipeline
  8. Parent Agent
  9. Planning Module.
  10. Child Agent
  11. Engagement Score Module.
  12. Moderator Agent
  13. Hierarchical Diversification Mechanism
  14. Data Level.
  15. Architectural Level.
  16. ...and 24 more sections

Figures (4)

  • Figure 1: SafeTalkCoach vs. Existing Frameworks
  • Figure 2: SafeTalkCoach's Multi-Agent Dialogue Generation Pipeline
  • Figure 3: Heuristic-based Engagement Score Module
  • Figure 4: Distribution of our Dataset