Exploding Mittens - Getting to grips with huge SKOS datasets

The ICONCLASS classification system contains more than 1 million SKOS nodes in English, German, French, Italian, Portuguese, Japanese, Dutch, Polish and Czech- in short a non-trivial large corpus of Linked Data. It is possible to load the data into performant triplestores, but becomes challenging in a production setting.Query performance using conventional SPARQL constructs is unacceptable to end-users, and finding relevant search results for users queries is an informational retrieval challenge due to the redundant yet slightly variant nature of the Art Historical classification data. We created a virtual Linked Data query engine that integrates a SPARQL API with semantic embeddings using BERT based transformers and RDF2Vec knowledge graph embeddings to provide performant queries in the Dutch Netwerk Digitaal Erfgoed termen netwerk.