Hi,
I have a question about the meta-path function for heterogeneous graphs. Does that meta-path function support the weighted graph? or Do you have the plan to develop a weighted meta-path function?
What kind of weighted meta-path functions do you want?
Hi, the method of this paper ’ Weighted Random Walks for Meta-Path Expansion in Heterogeneous Networks’.
Basically you want to sample edges based on their weights for some edge types in random-walk meta-path generation. I believe DGL has not supported something like this.
OK. Thank you. BTW, Do you have the plan to develop such kinds of APIs to support the weighted graph better?
I checked the paper you mentioned. Based on my understanding, the random walk is not about a “weighted metapath”. Rather, it is just a normal random walk based on a given metapath, with uniform transition probability on all other edge types except the “user-item” edge type.
In this case, dgl.random_walk already supports metapath-based random walk with non-uniform transition probabilities.
For weighted random walk that generates meta path, you will need to assign transition probabilities to the metagraph itself and use dgl.random_walk to generate the said metapath.
Thank you. Yes. That’s what I need. I did not know dgl has embedded the API of random_walk.