結構性對話，階層性行動：LLM多智能體系統的協作框架

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

February 16, 2025

作者: Zhao Wang, Sota Moriyama, Wei-Yao Wang, Briti Gangopadhyay, Shingo Takamatsu

cs.AI

摘要

最近在基於LLM的多智能體（LLM-MA）系統方面取得了一些進展，但在智能體協作處理複雜任務時，仍存在著重大挑戰，特別是在管理溝通和改進方面。本文提出了一個名為「結構化對話，階層化行動」（TalkHier）的新框架，該框架引入了一個結構化通訊協議，用於豐富上下文的交流，以及一個階層化改進系統，以應對不正確的輸出、虛假信息和偏見等問題。TalkHier在各種任務上超越了各種類型的最先進技術，包括推理擴展模型（OpenAI-o1）、開源多智能體模型（例如AgentVerse）和當前LLM和單智能體基準（例如ReAct、GPT4o）上的多數投票策略，這些任務包括開放領域問答、特定領域的選擇性提問和實用廣告文本生成。這些結果突顯了其為LLM-MA系統設定新標準的潛力，為更有效、適應性強且協作性強的多智能體框架鋪平了道路。代碼可在https://github.com/sony/talkhier找到。

English

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 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. 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.