Towards Long Context Hallucination Detection

2025-04-28

AI Summary

To address context hallucination—i.e., the generation of contradictory or unsupported content by large language models (LLMs) when processing long-context inputs—this paper introduces the first dedicated hallucination detection dataset for long-context scenarios and proposes a lightweight, decomposition-aggregation-based detection architecture. The architecture uniquely adapts pretrained encoders (e.g., BERT) to ultra-long text modeling via chunked encoding, cross-chunk aggregation, and contrastive hallucination discrimination learning—achieving high detection accuracy without compromising inference efficiency. Experiments demonstrate that the proposed method significantly outperforms same-scale baselines and LLM-based detectors across multiple metrics, while accelerating inference by several-fold. These results validate its effectiveness and practicality for long-context hallucination detection.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large Language Models (LLMs) have demonstrated remarkable performance across various tasks. However, they are prone to contextual hallucination, generating information that is either unsubstantiated or contradictory to the given context. Although many studies have investigated contextual hallucinations in LLMs, addressing them in long-context inputs remains an open problem. In this work, we take an initial step toward solving this problem by constructing a dataset specifically designed for long-context hallucination detection. Furthermore, we propose a novel architecture that enables pre-trained encoder models, such as BERT, to process long contexts and effectively detect contextual hallucinations through a decomposition and aggregation mechanism. Our experimental results show that the proposed architecture significantly outperforms previous models of similar size as well as LLM-based models across various metrics, while providing substantially faster inference.

Problem

• Detecting contextual hallucination in long-context LLM inputs
• Addressing unsubstantiated or contradictory LLM-generated information
• Enhancing hallucination detection efficiency and accuracy in long texts

Innovation

• Constructs dataset for long-context hallucination detection
• Proposes decomposition and aggregation architecture
• Enables BERT to process long contexts effectively