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

2026-06-02 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

A new training-free mixture-of-agents framework addresses challenges in Multi-Document Summarization (MDS), such as capturing complex inter-document relationships and reducing reliance on labeled data. This framework integrates Large Language Models (LLMs) and knowledge graphs to distill information from text collections. It decomposes the summarization process into three specialized agent tasks: extractive selection, knowledge-aware abstraction, and iterative refinement, all operating without task-specific fine-tuning. A multi-perspective consistency mechanism, guided by LLMs, unifies the outputs from these agents. Experiments conducted across four datasets in both English and Vietnamese languages demonstrate that the modular design achieves state-of-the-art or competitive performance, validating its effectiveness and adaptability for MDS tasks.

Key takeaway

For NLP Engineers developing multi-document summarization systems, this training-free agent framework offers a viable path to overcome data dependency and generalization issues. You should consider adopting a modular, agent-based design that integrates LLMs and knowledge graphs, especially if labeled data is scarce or domain adaptability is critical. This approach allows for state-of-the-art performance across diverse languages like English and Vietnamese without extensive fine-tuning.

Key insights

A training-free agent framework combines LLMs and knowledge graphs for effective multi-document summarization.

Principles

• Decompose complex tasks into specialized agents.
• Integrate knowledge graphs for relationship capture.

Method

The framework uses extractive selection, knowledge-aware abstraction, and iterative refinement agents, unifying their outputs via an LLM-guided multi-perspective consistency mechanism.

In practice

• Apply agent-based design to multi-document summarization.
• Utilize LLMs and KGs in training-free setups.

Topics

• Multi-Document Summarization
• Large Language Models
• Knowledge Graphs
• Agent-Based Systems
• Training-Free Models
• Natural Language Processing

Best for: AI Engineer, Machine Learning Engineer, Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Summarisation and Knowledge Distillation — How Agents Summarise Without Hallucinating — Robust LLM summarization requires architectural patterns to prevent hallucinations and ensure factual grounding and completeness.
• A Multimodal Text- and Graph-Based Approach for Open-Domain Event Extraction from Documents — MODEE combines LLMs and graph learning for robust open-domain event extraction, outperforming prior methods.
• AgREE: Agentic Reasoning for Knowledge Graph Completion on Emerging Entities — AgREE uses agent-based reasoning and iterative retrieval to complete knowledge graphs for emerging entities without training.
• Context Graphs: Free Association, Consistency, Narratives & Truth — LLMs function via free association on narrative graphs, trading consistency for broader, stochastically generated outputs.
• Leveraging LLM Parametric Knowledge for Fact Checking without Retrieval — Fact-checking without retrieval leverages LLM internal representations for robust, scalable claim verification.
• : Structuring Your Natural Language SOPs into Tailored Ambiguity-Resolved Code Templates — A multi-agent LLM framework effectively transforms ambiguous SOPs into structured, executable code with high accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.