Graph embeddings

In machine learning, a graph embedding is either

  1. an embedding of the nodes of a graph in Euclidean space that approximately preserves the graph’s structure, or
  2. an embedding of many whole graphs as single points in Euclidean space.

For topological graph theorists, “graph embedding” has a different meaning . Graph embeddings in ML have been inspired by the success of word embeddings.

Methods

  • LINE (2015): Large-scale Information Network Embedding (pdf, arxiv)
  • HOPE (2016): Higher-Order Proximity preserved Embedding (doi, pdf)
  • SDNE (2016): Structural Deep Network Embedding (doi, pdf)
  • HARP (2018): Hierarchical Representation Learning for Networks (pdf, arxiv)

Literature

Surveys

  • Goyal & Ferrara, 2018: Graph embedding techniques, applications, and performance: A survey (doi, arxiv)
  • Hamilton, Ying, Leskovec, 2017: Representation learning on graphs: Methods and applications (pdf, arxiv)
  • Zhang et al, 2018: Network representation learning: A survey (doi, arxiv)