Not All Claims Are Equally Risky: FACTOR for Adaptive Verification in Factual Long-Form Generation

2026-06-21

AI Summary

This work addresses the challenge of factual hallucinations in long-form text generation by large language models, where existing verification approaches apply uniform strategies to all claims despite varying risk levels. The authors propose FACTOR, a model-agnostic, adaptive verification framework that introduces claim-level risk awareness during inference. By estimating uncertainty to dynamically identify high-risk claims, FACTOR strategically allocates verification resources through adaptive language reasoning, candidate re-ranking, and targeted external evidence retrieval. Evaluated on the FactScore benchmark, the method significantly improves factual accuracy while reducing verification overhead. Ablation studies confirm the contribution of each component to the overall effectiveness of the framework.

Abstract

Large Language Models (LLMs) generate fluent long-form text, however, often add unsupported factual claims. Existing verification techniques improve factuality by grounding generation in external evidence. However, the same verification policy usually applies to all claims despite being differences in hallucination risks. We propose \textit{FACTOR} (\textit{FACTuality-Oriented Risk-aware Verification}), an inference-time model that adapts verification criteria according to claim-level uncertainty. FACTOR combines uncertainty estimation, adaptive language inference verification, and candidate re-ranking to allocate verification effort where it is most needed. We evaluate \textit{FACTOR} on FactScore benchmark showing that adaptive verification improves factuality while reducing verification cost simultaneously. We further perform different ablation studies to identify the primary driver of these gains. Our results show the effective and model-agnostic performance of \textit{FACTOR} for improving factuality in long-form generation.

Problem

• factuality
• hallucination risk
• adaptive verification
• long-form generation
• claim-level uncertainty

Innovation

• adaptive verification
• uncertainty estimation
• factuality
• long-form generation
• risk-aware