Random walk-based embeddings (DeepWalk, 2014; node2vec, 2016) generate node sequences by random walks, then feed them into Word2Vec's skip-gram model: nearby nodes in walks should have similar embeddings. The return bias p controls backward transitions — low p keeps the walk local (BFS-like), high p explores outward (DFS-like). The embedding space (right panel) projects 2D: nodes visited together cluster together. Visit frequency determines node brightness — well-connected hub nodes glow brightest as they appear most in walks.