Recovering latent cultural structure from intangible heritage via hybrid lexical–metadata clustering and similarity-based retrieval

Intangible cultural heritage (ICH) is a high-dimensional cultural knowledge space that remains largely inaccessible to computational design workflows due to fragmented representation and non-scalable manual curation. This study proposes a data-driven framework that integrates term frequency–inverse document frequency textual features with structured metadata to support unsupervised clustering and content-based recommendation of ICH elements. Using 500 United Nations Educational, Scientific and Cultural Organization (UNESCO)-referenced entries, a hybrid feature space was constructed and clustered using K-means (k = 4), with a silhouette score of 0.304, yielding interpretable cultural structure aligned with heritage typologies. Principal component analysis projections, distributional analyses, and similarity graphs demonstrate stable separation and semantically coherent retrieval. The results show that modest lexical signals, when fused with categorical metadata, are sufficient to recover interpretable, data-driven approximations of cultural structure and support similarity-based ideation. This work provides an interpretable, scalable, and domain-portable machine learning pipeline that advances computational cultural informatics and supports downstream creative tasks where cultural grounding is required.