CODEWITHSUNDEEP
graphartificial-intelligenceconv-neural-network
user10680899
user10680899

Reputation:

Node2Vec embeddings to global graph features

I want to extract global features for input graphs. I thought about using node2vec to build embeddings for each node of a graph. How can I use these node embeddings to come up with global features for the graphs? Does averaging the embeddings of all the nodes in a graph and considering the resulting vector the feature vector of the graph seem appropriate?

Upvotes: 5

Views: 457

Answers (1)

bluesummers
bluesummers

Reputation: 12627

I would consider two distinct approaches.

  1. Use node2vec, and use the node's embedding with document embedding methods from the NLP words to extract global features (Doc2vec etc).
  2. Use convolution network on top of a graph's edge matrix. Good post about that by Thomas Kipf.

Upvotes: 2

Related Questions