FactGuard: Leveraging Multi-Agent Systems to Generate Answerable and Unanswerable Questions for Enhanced Long-Context LLM Extraction

2025-04-08

AI Summary

Existing long-context extractive QA models exhibit weak refusal capability and poor factual consistency. Method: We propose a multi-agent collaborative data augmentation framework wherein specialized agents—performing fact extraction, question generation, and contradiction verification—cooperate to automatically synthesize high-quality answerable/unalanswerable question pairs over long contexts (8K–128K tokens). Our approach integrates evidence-chain-guided unanswerable question synthesis and context-aware hard-example mining, eliminating reliance on manual annotation. Contribution/Results: This is the first work enabling large-scale, controllable hard-example generation without human labeling—filling a critical gap since SQuAD 2.0. We introduce FactGuard-Bench, a benchmark of 25,220 instances, on which seven mainstream LLMs achieve only 61.79% average accuracy, underscoring the severity of the problem. Our method significantly improves refusal robustness and factual consistency, establishing a new paradigm for trustworthy training of long-context LLMs.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Multiagent Systems: Adversarial Agents
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Extractive reading comprehension systems are designed to locate the correct answer to a question within a given text. However, a persistent challenge lies in ensuring these models maintain high accuracy in answering questions while reliably recognizing unanswerable queries. Despite significant advances in large language models (LLMs) for reading comprehension, this issue remains critical, particularly as the length of supported contexts continues to expand. To address this challenge, we propose an innovative data augmentation methodology grounded in a multi-agent collaborative framework. Unlike traditional methods, such as the costly human annotation process required for datasets like SQuAD 2.0, our method autonomously generates evidence-based question-answer pairs and systematically constructs unanswerable questions. Using this methodology, we developed the FactGuard-Bench dataset, which comprises 25,220 examples of both answerable and unanswerable question scenarios, with context lengths ranging from 8K to 128K. Experimental evaluations conducted on seven popular LLMs reveal that even the most advanced models achieve only 61.79% overall accuracy. Furthermore, we emphasize the importance of a model's ability to reason about unanswerable questions to avoid generating plausible but incorrect answers. By implementing efficient data selection and generation within the multi-agent collaborative framework, our method significantly reduces the traditionally high costs associated with manual annotation and provides valuable insights for the training and optimization of LLMs.

Problem

• Improving accuracy in distinguishing answerable and unanswerable questions for LLMs
• Reducing manual annotation costs for long-context question-answer datasets
• Enhancing LLM reasoning to avoid plausible but incorrect answers

Innovation

• Multi-agent system generates question-answer pairs
• Autonomously constructs unanswerable questions systematically
• Reduces manual annotation costs efficiently