Biased random walks, skip-gram embedding, 2D t-SNE visualization of node communities
1.0
1.0
15
8
node2vec (Grover & Leskovec, 2016): generates node embeddings via biased random walks. Parameter p controls return probability (BFS-like when p small), and q controls exploration (DFS-like when q small). The walk generates context pairs for skip-gram training (Word2Vec objective). Communities cluster in embedding space. t-SNE compresses high-dimensional embeddings to 2D while preserving local structure, revealing structural roles and community membership.