Native Multimodal Knowledge Graph Construction Using Hypergraph Structures for Enhanced Retrieval-Augmented Generation Systems

This repository releases the code implementation and accompanying dataset of a multimodal hypergraph corpus. The dataset comprises annotated documents featuring 8,432 entities, 3,276 n-ary hyperedge relations with temporal and spatial grounding, and 2,847 cross-modal coreference links spanning text, image, audio, and video modalities. It serves as a benchmark resource for research on hypergraph-based multimodal knowledge graphs and retrieval-augmented generation systems.