Representation Learning for the Semantic Web

In the past years, learning vector space embeddings has rapidly gained attention, first in the natural language processing community with the advent of word2vec, and more recently also in the Semantic Web community, eg, with the adaptations RDF2vec or node2vec, as well as the RESCAL, HolE and Trans* family. Their properties-the representation of entities in a dense vector space, the proximity of semantically related entities, and the preservation of the direction of semantic relations-make them interesting for many applications. Consequently, the field of embedding learning has recently gained a considerable uptake in the Semantic Web community.There are various ways of creating such embeddings. They range from applying the word2vec paradigm to sequences derived from graphs to translation learning and tensor factorization. Those methods differ in many aspects, such as the strategies used, the …