Talk Structurally, Act Hierarchically: A Collaborative Framework for LLM Multi-Agent Systems

2025-02-16

AI Summary

To address critical challenges in LLM-based multi-agent collaboration—including chaotic communication, frequent output errors, and pronounced bias/hallucination—this paper proposes a dual-track paradigm integrating *dialogue structuring* and *action layering*. Methodologically, we design a lightweight structured communication protocol enabling context-aware, precise agent interaction; construct a hierarchical reflection-refinement system coupled with a multi-stage consensus verification mechanism to support progressive output correction; and introduce a task-adaptive role-allocation strategy to enhance collaborative efficiency. Evaluated on three diverse tasks—open-domain question answering, domain-selective question posing, and advertising copy generation—our approach significantly outperforms state-of-the-art baselines (e.g., OpenAI-o1, AgentVerse, ReAct), markedly reducing hallucinations and bias while improving output reliability and factual consistency. The implementation is publicly available.

Technology Category

• Multiagent Systems: Agent Communication
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Recent advancements in LLM-based multi-agent (LLM-MA) systems have shown promise, yet significant challenges remain in managing communication and refinement when agents collaborate on complex tasks. In this paper, we propose extit{Talk Structurally, Act Hierarchically (TalkHier)}, a novel framework that introduces a structured communication protocol for context-rich exchanges and a hierarchical refinement system to address issues such as incorrect outputs, falsehoods, and biases. extit{TalkHier} surpasses various types of SoTA, including inference scaling model (OpenAI-o1), open-source multi-agent models (e.g., AgentVerse), and majority voting strategies on current LLM and single-agent baselines (e.g., ReAct, GPT4o), across diverse tasks, including open-domain question answering, domain-specific selective questioning, and practical advertisement text generation. These results highlight its potential to set a new standard for LLM-MA systems, paving the way for more effective, adaptable, and collaborative multi-agent frameworks. The code is available https://github.com/sony/talkhier.

Problem

• Improve communication in multi-agent systems
• Enhance task refinement for accuracy
• Reduce biases in collaborative AI frameworks

Innovation

• Structured communication protocol
• Hierarchical refinement system
• Surpasses SoTA models