Uniform walks

Simple random walk (no biasing). Fixed walk length, multiple walks per node.

Advertisement

Skip-gram

Same objective as word2vec. Walks = 'sentences'. Nodes = 'words'.

Advertisement

Hierarchical softmax

Efficient training via binary tree over nodes. O(log V) per prediction instead of O(V).

vs node2vec

DeepWalk uniform, node2vec biased. node2vec generalizes DeepWalk.