Pytorch Geometric tutorial: DeepWalk and Node2Vec (Theory)

We present the idea of using language models and adapt them to the graph setting by means of random walks sampling. In particular, we discuss the similarity and difference of the two approaches, by highlighting the fundamental ideas introduced by DeepWalk, and the generalizations provided by node2vec.

A second tutorial (next week) will present the computational details of the methods (i.e., Hierarchical Softmax and Negative Sampling), and discuss the implementation of the methods in PyTorch.

Download the material from our official website:

[truth_seeker]

I think the slide shown at 21:05 is incorrect. BFS helps with homophily/communities while DFS helps with structural equivalence.

From the original paper:

"For instance, in Figure 1, we observe nodes u and s1 belonging to the same tightly knit community of nodes,
while the nodes u and s6 in the two distinct communities share the same structural role of a hub node"

[VLM]

Hi,
1. How the embedding function(f) is mapping a node(feature representation) into a "d" dimensional vector, is there any matrix if so how to initialize it?
2. How is aggregation being done after the random walk?

[yangwang]

Hello, thanks for the tutorial very much. In the BFS, DFS example, the DFS first will make a u, S4, S5, S6 path and catch similarity between u and S6. My question is, in this DFS path, is the embedding of u and S5 are more close than u and S6? u and S6 are kind of the center of their clusters, could the DFS path catch this information?

[krishnakantsingh]

Hi,
Would it be possible to have (at-least) one tutorial on model interpretability.
Thank you for making this series of videos.

[zahrapoorsoltani]

voice quality is awful and hard to understand