The Future of Cognitive Strategy-enhanced Persuasive Dialogue Agents: New Perspectives and Trends

Mengqi Chen, Bin Guo, Hao Wang, Haoyu Li, Qian Zhao, Jingqi Liu, Yasan Ding, Yan Pan, Zhiwen Yu

Abstract

Persuasion, as one of the crucial abilities in human communication, has garnered extensive attention from researchers within the field of intelligent dialogue systems. We humans tend to persuade others to change their viewpoints, attitudes or behaviors through conversations in various scenarios (e.g., persuasion for social good, arguing in online platforms). Developing dialogue agents that can persuade others to accept certain standpoints is essential to achieving truly intelligent and anthropomorphic dialogue system. Benefiting from the substantial progress of Large Language Models (LLMs), dialogue agents have acquired an exceptional capability in context understanding and response generation. However, as a typical and complicated cognitive psychological system, persuasive dialogue agents also require knowledge from the domain of cognitive psychology to attain a level of human-like persuasion. Consequently, the cognitive strategy-enhanced persuasive dialogue agent (defined as CogAgent), which incorporates cognitive strategies to achieve persuasive targets through conversation, has become a predominant research paradigm. To depict the research trends of CogAgent, in this paper, we first present several fundamental cognitive psychology theories and give the formalized definition of three typical cognitive strategies, including the persuasion strategy, the topic path planning strategy, and the argument structure prediction strategy. Then we propose a new system architecture by incorporating the formalized definition to lay the foundation of CogAgent. Representative works are detailed and investigated according to the combined cognitive strategy, followed by the summary of authoritative benchmarks and evaluation metrics. Finally, we summarize our insights on open issues and future directions of CogAgent for upcoming researchers.

Introduction

While Large Language Models (LLMs) like ChatGPT are incredibly fluent, they often lack the strategic depth required for effective persuasion. Persuasion isn't just about being eloquent; it’s a complex psychological process involving "Theory of Mind" (understanding what the other person is thinking), emotional resonance, and logical consistency.

Current dialogue agents often struggle with long-term planning, psychological grounding, and evaluation. They might win a single turn but lose the overall "argument" by failing to guide the conversation toward a specific goal. Furthermore, standard metrics like BLEU or ROUGE measure text similarity rather than the actual psychological impact or change in a user's stance.

The CogAgent Framework

The authors propose a unified framework for CogAgent based on three core cognitive pillars designed to move beyond reactive chitchat toward goal-oriented influence.

1. Persuasion Strategy

This involves selecting specific linguistic techniques based on the user's current state. The paper formalizes a taxonomy of strategies, such as "Present of Facts" (logical appeal), "Emotional Resonance" (pathos), and "Self-modeling" (acting as a role model).

A taxonomy of persuasion strategies used by CogAgents.

2. Topic Path Planning Strategy

Instead of reactive chatting, the agent proactively plans a sequence of topics to navigate the user from their current state to a target standpoint. For example, an agent might move a user from being "anxious about money" to being "willing to seek professional counseling" by first validating their emotions and then gradually introducing solutions.

Example of a persuasive dialogue for psychological counseling illustrating topic path planning.

3. Argument Structure Prediction Strategy

The agent mines and predicts authoritative arguments and counter-arguments. This ensures the agent's claims are plausible and logically structured, often using graph-based models to represent the relationships between different points of view.

Key Insights and Strategic Ideas

  • The "Pre-suasion" Factor: Borrowing from Robert Cialdini, the authors emphasize that when you say something is as important as what you say. Effective agents must prime the user before making a persuasive request.
  • Theory of Mind (ToM) as a Feature: Successful persuasion requires the agent to maintain a mental model of the user's beliefs and emotions. The paper highlights works that explicitly model ToM to predict user reactions.
  • From Reactive to Proactive: The shift from "chitchat" to "target-guided" dialogue is the defining characteristic of CogAgent. This requires moving from simple sequence-to-sequence models to Reinforcement Learning (RL) and graph-based planning.
  • The Logic-Emotion Balance: Drawing on Aristotle, the paper notes that while logic (logos) is essential for credibility, emotion (pathos) is often what actually triggers a change in decision-making.

Datasets and Performance

The paper reviews datasets across five key persuasive scenarios, including Psychological Counseling (ESConv), Debate (Winning Arguments), Price Negotiation (CraigslistBargain), Donation (Persuasion For Good), and Product Recommendation (TG-ReDial).

Research indicates that systems explicitly planning persuasion strategies consistently outperform vanilla models in both human preference and task success rates. Recent studies show that LLMs, when fine-tuned or prompted with these cognitive strategies, achieve a level of logic and fluency previously impossible with smaller, specialized models. The most effective systems currently combine the structured reasoning of graphs for topic planning with the generative power of LLMs.

Limitations and Future Work

The authors identify several "open frontiers" for the field:

  • The Evaluation Gap: There is still no "gold standard" automatic metric for persuasion. Developing metrics that account for psychological impact is a high priority.
  • Multi-party Persuasion: Most current research focuses on 1-on-1 conversations. Real-world persuasion often happens in groups, requiring agents to manage complex social dynamics.
  • Multimodality: Humans persuade using tone of voice, facial expressions, and gestures. Future CogAgents need to integrate visual and auditory cues.
  • Interpretability: For high-stakes applications like counseling, we need to understand why an agent chose a specific strategy, rather than treating the LLM as a black box.

Summary updated

Abstract

Persuasion, as one of the crucial abilities in human communication, has garnered extensive attention from researchers within the field of intelligent dialogue systems. We humans tend to persuade others to change their viewpoints, attitudes or behaviors through conversations in various scenarios (e.g., persuasion for social good, arguing in online platforms). Developing dialogue agents that can persuade others to accept certain standpoints is essential to achieving truly intelligent and anthropomorphic dialogue system. Benefiting from the substantial progress of Large Language Models (LLMs), dialogue agents have acquired an exceptional capability in context understanding and response generation. However, as a typical and complicated cognitive psychological system, persuasive dialogue agents also require knowledge from the domain of cognitive psychology to attain a level of human-like persuasion. Consequently, the cognitive strategy-enhanced persuasive dialogue agent (defined as CogAgent), which incorporates cognitive strategies to achieve persuasive targets through conversation, has become a predominant research paradigm. To depict the research trends of CogAgent, in this paper, we first present several fundamental cognitive psychology theories and give the formalized definition of three typical cognitive strategies, including the persuasion strategy, the topic path planning strategy, and the argument structure prediction strategy. Then we propose a new system architecture by incorporating the formalized definition to lay the foundation of CogAgent. Representative works are detailed and investigated according to the combined cognitive strategy, followed by the summary of authoritative benchmarks and evaluation metrics. Finally, we summarize our insights on open issues and future directions of CogAgent for upcoming researchers.
BibTeX
@article{chen2025future,
  author = {Mengqi Chen and Bin Guo and Hao Wang and Haoyu Li and Qian Zhao and Jingqi Liu and Yasan Ding and Yan Pan and Zhiwen Yu},
  title = {{The Future of Cognitive Strategy-enhanced Persuasive Dialogue Agents: New Perspectives and Trends}},
  journal = {Frontiers of Computer Science},
  year = {2025},
  url = {https://openreview.net/forum?id=LQnSo9ioil},
}
Figure