Scholarlink

<p>Large Language Models (LLMs) often struggle with multi-step reasoning tasks. To address this, LLM-based Multi-Agent Systems (LLM-MAS) have been introduced, where agents interact to reach consensus. However, these systems face two major issues: hallucination propagation, where incorrect responses mislead others, and communication redundancy, where excessive interactions increase token cost without improving performance. This paper proposes a unified solution: Consensus-driven Community Agent Communication (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\text {AC}^{\varvec{3}}\)</EquationSource> </InlineEquation>). Each agent’s response is decomposed into an answer and an analysis, allowing for precise tracking of misleading content and redundant patterns. Agents are clustered into communities based on answers, and selected representatives—chosen by analysis quality—conduct cross-community discussion. A confirmation mechanism finalizes the consensus. Experiments on multiple datasets show that <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\text {AC}^{\varvec{3}}\)</EquationSource> </InlineEquation> significantly reduces hallucinations and communication overhead while maintaining or improving reasoning accuracy. These results demonstrate the effectiveness and scalability of <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\text {AC}^{\varvec{3}}\)</EquationSource> </InlineEquation> in enhancing LLM-based multi-agent systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic Consensus Communication Mechanism for Large Language Model-Based Multi-Agent Systems

  • 期刊论文

  • 发表日期：

    2026年1月14日

查看全文
  • Liancheng Yang,
  • Sining Li,
  • Aobo Deng

摘要

Large Language Models (LLMs) often struggle with multi-step reasoning tasks. To address this, LLM-based Multi-Agent Systems (LLM-MAS) have been introduced, where agents interact to reach consensus. However, these systems face two major issues: hallucination propagation, where incorrect responses mislead others, and communication redundancy, where excessive interactions increase token cost without improving performance. This paper proposes a unified solution: Consensus-driven Community Agent Communication ( \(\text {AC}^{\varvec{3}}\) ). Each agent’s response is decomposed into an answer and an analysis, allowing for precise tracking of misleading content and redundant patterns. Agents are clustered into communities based on answers, and selected representatives—chosen by analysis quality—conduct cross-community discussion. A confirmation mechanism finalizes the consensus. Experiments on multiple datasets show that \(\text {AC}^{\varvec{3}}\) significantly reduces hallucinations and communication overhead while maintaining or improving reasoning accuracy. These results demonstrate the effectiveness and scalability of \(\text {AC}^{\varvec{3}}\) in enhancing LLM-based multi-agent systems.