Intangible cultural heritage (ICH) archives face mounting challenges in intelligent organisation, semantic integration, and innovative transmission. ICH materials are inherently performative, contextual, and tacit, and are scattered across heterogeneous formats including texts, images, audio-visual records, and three-dimensional representations. Existing digital archival practice prioritises static preservation over dynamic knowledge representation, leaving semantic depth shallow and support for cultural interpretation weak (Hou et al. 2022; Hawkins 2022). New knowledge-organisation paradigms are therefore needed, ones that combine artificial intelligence, multimodal data processing, and cultural semantics.

Knowledge graphs have emerged as a core infrastructure for structuring cultural heritage entities, relationships, and contextual meanings (Hyvönen 2020; Huang et al. 2023). Advances in natural language processing and computer vision have further enabled large-scale semantic extraction and annotation of heritage data (Abgaz et al. 2021). More recently, Artificial Intelligence Generated Content (AIGC), driven by large language models and multimodal generative models, has demonstrated strong capabilities in content generation, semantic enrichment, and cross-modal reasoning (Li et al. 2025; Qi / Wen 2025). Yet empirical research on how AIGC can be systematically integrated into archival knowledge-graph construction for ICH remains limited.

This study proposes an AIGC-enabled framework for constructing a multimodal knowledge graph of ICH, using Chinese paper-cutting as a representative case. Recognised by UNESCO as Intangible Cultural Heritage of Humanity, Chinese paper-cutting embodies rich symbolic meanings, regional variations, and craft knowledge transmitted primarily through practice and oral instruction, making it an ideal testbed for exploring how generative AI can articulate and structure implicit cultural knowledge.

Methodologically, the study introduces an integrated four-stage framework: collection–enhancement–organisation–narration. (1) Heterogeneous paper-cutting data, comprising textual descriptions, high-resolution images, audio-visual technique recordings, and three-dimensional models, are systematically collected and harmonised. (2) Large language models and multimodal generative models perform deep semantic analysis, automatic annotation, and content enhancement, generating derivative knowledge such as procedural technique descriptions, symbolic-motif interpretations, and regional-custom explanations. (3) Domain ontology modelling is used to extract core entities, including techniques, patterns, tools, inheritors, regions, and folk symbols, and to link them semantically into a multimodal knowledge graph with rich relational structures. (4) The graph is integrated with immersive technologies, including virtual reality, to support visualisation, dynamic storytelling, intelligent question answering, and interactive learning.

Empirical results demonstrate that the framework effectively integrates fragmented ICH knowledge, externalises tacit cultural logic, and enhances the narrative and interactive capacities of archival resources. Theoretically, this study extends the post-custodial archival paradigm toward datafication, semanticisation, and intelligent knowledge services, offering a scalable model for digital memory construction of ICH. Practically, it provides cultural heritage institutions with an operational pathway to shift from passive custodians to active memory constructors, supporting the creative transformation and innovative development of intangible cultural heritage in the era of artificial intelligence.