Hallucination Detection in Long-Form Text Generated by LLMs: A Benchmark and a Hyper-Relational Knowledge Graph Approach

Hallucination detection has received increasing attention, particularly in long-form text generation, where language models are more prone to producing factually inaccurate content. Previous studies reveal two limitations: (1) current benchmarks focus on short-form content, lacking the structural complexity required in long-form scenarios; (2) existing methods are constrained by coarse-grained consistency checks and fail to capture long-range and hyperrelational dependencies. To address these challenges, we provide LHD, a benchmark for longform hallucination detection that contains diverse entity types and intricate factual dependencies spanning extended contexts. We further propose HRKG-HD, a black-box framework that models responses as fact-centric hyperrelational knowledge graphs and detects hallucinations through relation-aware multi-hop reasoning over these graphs. By linking distant facts through shared entities and qualifiers, this design enables a global and dependency-aware verification of factual consistency. Extensive experiments demonstrate that HRKG-HD not only outperforms existing baselines at both the passage-level and sentence-level, but also exhibits robust and consistent performance across various LLMs. The code and dataset will be released upon acceptance to facilitate reproducibility.