Parallelism Meets Adaptiveness: Scalable Documents Understanding in Multi-Agent LLM Systems

2025-07-22

AI Summary

Existing multi-agent frameworks for complex document understanding suffer from static workflows, fixed agent roles, and inefficient inter-agent communication. To address these limitations, this paper proposes a dynamic collaborative large language model (LLM) multi-agent system. Our method introduces three core innovations: (1) confidence-driven dynamic task routing and reallocation; (2) structured bidirectional critique exchange coupled with parallel agent evaluation; and (3) a competitive result selection strategy guided by multi-dimensional quality criteria. The system adopts a modular architecture integrating LLM-based agents, structured feedback mechanisms, and adaptive coordination control. Experimental results demonstrate significant improvements over static and partially adaptive baselines across factual coverage, content coherence, and execution efficiency—validating the effectiveness of the “dynamic collaboration + structured competition” paradigm.

Technology Category

• Multiagent Systems: Multiagent Learning
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language model (LLM) agents have shown increasing promise for collaborative task completion. However, existing multi-agent frameworks often rely on static workflows, fixed roles, and limited inter-agent communication, reducing their effectiveness in open-ended, high-complexity domains. This paper proposes a coordination framework that enables adaptiveness through three core mechanisms: dynamic task routing, bidirectional feedback, and parallel agent evaluation. The framework allows agents to reallocate tasks based on confidence and workload, exchange structured critiques to iteratively improve outputs, and crucially compete on high-ambiguity subtasks with evaluator-driven selection of the most suitable result. We instantiate these principles in a modular architecture and demonstrate substantial improvements in factual coverage, coherence, and efficiency over static and partially adaptive baselines. Our findings highlight the benefits of incorporating both adaptiveness and structured competition in multi-agent LLM systems.

Problem

• Static workflows limit multi-agent LLM effectiveness
• Open-ended tasks need dynamic task routing
• High-ambiguity subtasks require evaluator-driven competition

Innovation

• Dynamic task routing based on confidence
• Bidirectional feedback for iterative improvement
• Parallel agent evaluation with competition