SMADE-IE: Sparse Multi-Agent Framework with Evidence-Driven Debate for Zero-Shot Information Extraction

2026-06-03

AI Summary

This work addresses key limitations in zero-shot information extraction, where monolithic prompting often yields boundary and type errors, while per-type prompting or multi-agent debate incurs cross-type conflicts, redundant interactions, and high computational costs. To overcome these issues, the authors propose a sparse, evidence-driven multi-agent framework that dynamically switches between global and type-centric extraction strategies via an adaptive schema selector. Conflicting predictions are resolved through a Toulmin-structure-based, evidence-driven debate mechanism, augmented by external evidence scoring and Bayesian confidence updating. Evaluated across nine benchmark datasets for named entity recognition, relation extraction, and joint entity and relation extraction, the method significantly outperforms existing zero-shot approaches while achieving substantial token efficiency gains through sparse agent scheduling and early stopping.

Abstract

Zero-shot information extraction (IE) with large language models (LLMs) has attracted increasing attention due to its flexibility in adapting to new schemas and domains without task-specific training. Existing approaches mainly rely on monolithic prompting, each-type prompting, or multi-agent debate. However, monolithic prompting often suffers from boundary and type errors, while each-type prompting and multi-agent debate introduce cross-type conflicts, redundant agent interactions, and substantial token overhead. To address these challenges, we propose SMADE-IE, a sparse and evidence-driven multi-agent framework for zero-shot IE. SMADE-IE first employs an Adaptive Mode Selector to dynamically route inputs into either a lightweight Global Extraction Mode or a Type-Centric Extraction Mode, reducing unnecessary type selection and reasoning noise. For conflicting predictions, we further introduce an Evidence-Driven Debate mechanism that structures arguments into Toulmin-style components and performs confidence aggregation through external evidence scoring and Bayesian updates. Experimental results on 9 benchmark datasets across NER, RE, and JERE tasks show that SMADE-IE consistently outperforms existing zero-shot IE baselines while also improving token efficiency through sparse agent selection and early-stopping debate.

Problem

• zero-shot information extraction
• large language models
• multi-agent debate
• token overhead
• cross-type conflicts

Innovation

• Sparse Multi-Agent
• Evidence-Driven Debate
• Zero-Shot Information Extraction
• Adaptive Mode Selection
• Toulmin-style Argumentation