A Training-Free Mixture-of-Agents Framework for Multi-Document Summarization using LLMs and Knowledge Graphs

2026-06-02

AI Summary

This work addresses key challenges in multi-document summarization—namely, the difficulty of modeling inter-document relationships, reliance on large annotated datasets, and limited cross-domain and cross-lingual generalization. The authors propose a novel, training-free hybrid agent framework that integrates large language models with knowledge graphs, decomposing the task into three stages: extractive selection, knowledge-aware abstraction, and iterative refinement. A multi-perspective consistency mechanism is employed to fuse outputs from these stages. This approach establishes the first fine-tuning-free multi-agent collaborative architecture for summarization, substantially enhancing semantic comprehension and adaptability across languages and domains. Evaluated on four English–Vietnamese benchmark datasets, the method achieves state-of-the-art or competitive performance, demonstrating its effectiveness and broad applicability.

Abstract

Multi-Document Summarization (MDS) plays a critical role in distilling essential information from collections of textual data. Existing approaches often struggle to capture complex inter-document relationships, rely heavily on large amounts of labeled data for supervised training, or exhibit limited generalization across domains and languages. To address these limitations, we present a training-free mixture-of-agents framework for MDS that leverages the complementary strengths of large language models (LLMs) and knowledge graphs. Our approach decomposes summarization into specialized agent tasks: extractive selection, knowledge-aware abstraction, and iterative refinement, each operating without task-specific fine-tuning. We unify their outputs using a multi-perspective consistency mechanism guided by LLMs. Experiments across four datasets in English and Vietnamese demonstrate state-of-the-art or competitive performance, validating the effectiveness and adaptability of our modular design.

Problem

• Multi-Document Summarization
• inter-document relationships
• supervised training
• generalization
• large language models

Innovation

• training-free
• mixture-of-agents
• multi-document summarization
• knowledge graphs
• large language models