A scalable approach to incrementally building knowledge graphs

Abstract

We work on converting the metadata of 13 American art museums and archives into Linked Data, to be able to integrate and query the resulting data. While there are many good sources of artist data, no single source covers all artists. We thus address the challenge of building a comprehensive knowledge graph of artists that we can then use to link the data from each of the individual museums. We present a framework to construct and incrementally extend a knowledge graph, describe and evaluate techniques for efficiently building knowledge graphs through the use of the MinHash/LSH algorithm for generating candidate matches, and conduct an evaluation that demonstrates our approach can efficiently and accurately build a knowledge graph about artists.