Persuasion Should be Double-Blind: A Multi-Domain Dialogue Dataset With Faithfulness Based on Causal Theory of Mind

TL;DR

The paper addresses a key gap in persuasive dialogue research: real-world interactions require preserving double-blind information and inferring the persuadee’s mental state. It introduces ToMMA, a multi-agent framework that uses causal Theory of Mind to guide dialogue without leaking internal states, and presents CToMPersu, a large-scale, multi-domain dataset with 6,275 dialogues across 35 domains and 6,257 scenarios. The approach combines mental-state generation, constrained dialogue generation, and observer-based quality control to improve logical coherence and human-likeness beyond prior GPT-generated data. Empirical evaluation shows CToMPersu achieves better alignment with human dialogues, especially under Causal Theory of Mind Evaluation, while highlighting the continued need for ToM and ethics-aware development in persuasive AI.

Abstract

Persuasive dialogue plays a pivotal role in human communication, influencing various domains. Recent persuasive dialogue datasets often fail to align with real-world interpersonal interactions, leading to unfaithful representations. For instance, unrealistic scenarios may arise, such as when the persuadee explicitly instructs the persuader on which persuasion strategies to employ, with each of the persuadee's questions corresponding to a specific strategy for the persuader to follow. This issue can be attributed to a violation of the "Double Blind" condition, where critical information is fully shared between participants. In actual human interactions, however, key information such as the mental state of the persuadee and the persuasion strategies of the persuader is not directly accessible. The persuader must infer the persuadee's mental state using Theory of Mind capabilities and construct arguments that align with the persuadee's motivations. To address this gap, we introduce ToMMA, a novel multi-agent framework for dialogue generation that is guided by causal Theory of Mind. This framework ensures that information remains undisclosed between agents, preserving "double-blind" conditions, while causal ToM directs the persuader's reasoning, enhancing alignment with human-like persuasion dynamics. Consequently, we present CToMPersu, a multi-domain, multi-turn persuasive dialogue dataset that tackles both double-blind and logical coherence issues, demonstrating superior performance across multiple metrics and achieving better alignment with real human dialogues. Our dataset and prompts are available at https://github.com/DingyiZhang/ToMMA-CToMPersu .

Figures (5)

• Figure 1: An example illustrating the unnaturalness of an LLM-generated dataset. In the figure, the blue text highlights instances where the persuadee mistakenly adopts the persuader's arguments while expressing their own viewpoint. Moreover, as indicated by the red text, the persuadee never actively presents arguments supporting their presumed stance—in this case, the benefits of the Shopping Mall. Instead, they merely guide the persuader to apply persuasion techniques on them.
• Figure 2: Causal Theory of Mind
• Figure 3: Overview of the ToMMA framework for collecting the CToMPersu dataset. This figue illustrates the three-step process: (1) Mental State Generation, (2) Dialogue Generation Guided by Causal Theory of Mind, and (3) Observer Interaction for quality control.
• Figure 4: 3rd Round Persuader Response Prompt Design
• Figure 5: An example demonstrating the effectiveness of the observer agent. In this round, the persuader is supposed to address the desire. However, both the mental state prediction and the persuasive dialogue generation incorrectly focus too much on belief. In the end, the entire issue is resolved by the observer agent.