Dynamic Weighted Consensus Framework for LLM Multi-agent Debate

doi:10.1007/978-981-95-3001-4_14

Dynamic Weighted Consensus Framework for LLM Multi-agent Debate

2026-01-01

0

PRE

AI

DOI：10.1007/978-981-95-3001-4_14

Original

Original request for help

Save

Abstract

En 中文

Large Language Models (LLMs) have demonstrated remarkable capabilities in complex reasoning and fact verification tasks, yet their performance remains constrained by inherent uncertainties in multi-step reasoning processes. Although the Multi-Agent Debate (MAD) framework significantly enhances reasoning accuracy through collaborative interactions among agents, these agents often generate divergent answers to the same question, which may inadvertently propagate incorrect responses. In such scenarios, greater priority should be granted to agents that consistently provide accurate answers. To address this challenge, this paper proposes a novel Dynamic Weighted Consensus Framework for LLM Multi-Agent Debate (DWC-MAD). This framework integrates real-time confidence quantification (derived from agents' iterative responses) and longitudinal accuracy metrics (based on agents' historical performance). By employing a dynamic weight allocation algorithm, DWC-MAD assigns differentiated weights to agent inputs, thereby constructing an optimized multi-agent debate system.

Keywords:

LLMs

Multi-Agent Debate Framework

Dynamic Weight Allocation

Collaborative Reasoning

Uncertainty Quantification

Dynamic Weighted Consensus Framework for LLM Multi-agent Debate

Abstract

Journal

Organization