Data Science Desktop Survival Guide
by Graham Williams
A graph embedding learns a mapping from the graph to a more traditional vector space as utilised by data scientists, while preserving relevant network properties. Vector operations tend to be simpler and faster than the same operations on graphs. Traditional machine learning and statistics tend to operate in vector spaces.
A typical example is nearest neighbours. Within a graph space links can be traversed from node to node as we travel further from a node relationships become less meaningful. On transforming to vector space distance metrics can be used over the features for a more straightforward nearest neighbours.
See “Graph Embedding Techniques, Applications, and Performance: A Survey” from https://arxiv.org/abs/1705.02801.