Random walks
Two parameters: p (return), q (in-out). Balance BFS-like (structural) vs DFS-like (community). Neighborhood contexts.
Advertisement
Skip-gram objective
Predict context nodes given center. Learn embeddings maximizing likelihood.
Advertisement
Complexity
Preprocessing: O(V + E). Walk generation: O(walks × length). Training: O(iterations × walks × window).
Downstream tasks
Link prediction. Node classification. Similarity search. Any task benefiting from continuous features.