I have been playing around with MetaPath2Vec recently on a custom dataset. Couldn’t tell from documentation whether the native implementation (MetaPath2Vec — DGL 1.1.1 documentation) aggregates any edge data or node data on a heterogeneous graph in its embeddings. Could someone explain to me if it does, and if yes - how?
It learns the node embedding based on the graph’s topology only, so no edge data is aggregated into the learned embeddings.
Thank you for the response @dyru, couple of follow ups:
- Would it make sense to use edge weights as probabilities for traversing the meta-path random walks?
- Is there any unsupervised embeddings algorithm on DGL that could be used for weighted heterogeneous graphs?
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.