Node2Vec

The DeepWalk algorithm uses unbiased randomized walks to generate the neighborhood of any concerned node. Its unbiased nature ensures the graph structure is captured in the best possible manner statistically, but, in practice, this is often the less optimal choice. The premise of Node2Vec is that we introduce bias in the random walk strategy to ensure that sampling is done in such a way that both the local and global structures of the graph are represented in the neighborhood. Most of the other concepts in Node2Vec are the same as those for DeepWalk, including the learning objective and the optimization step.

Before we delve into the nitty-gritty of the algorithm, let’s do a quick recap of graph traversal approaches.

Graph traversal approaches

As we covered briefly in Chapter 1, the two most popular graph traversal approaches are breadth-first search (BFS) and depth-first search (DFS). BFS is the local first approach to graph exploration where, given a starting...