THUDM / GATNE

Source code and dataset for KDD 2019 paper "Representation Learning for Attributed Multiplex Heterogeneous Network"
MIT License
525 stars 141 forks source link

How features are generated? #110

Open geo47 opened 3 years ago

geo47 commented 3 years ago

Hello,

Could you please guide how you generated feature embeddings for each node?

More specifically, I guess the feature embedding represents the attribute features of the node. If we use Node2Vec for embedding it only gives the node embedding, however using GraphSAGE we can obtain embedding with node and attributes.

Thanks

cenyk1230 commented 3 years ago

Hi @geo47,

What do feature embeddings mean? Actually, our method is similar to unsupervised GraphSAGE to generate node embeddings with both structural and attribute information. But we mainly focus on the multiplex heterogeneous scenarios.

geo47 commented 3 years ago

Hi @cenyk1230 ,

Thanks for the reply.

By feature embedding, I mean how you generate the feature embedding in the feature.txt file.

Like, if I want to generate it on my own custom data, how can I do that as I need to generate node embeddings with both structural and attribute information.

Thanks.

cenyk1230 commented 3 years ago

Hi @geo47,

It depends on your dataset. Does your dataset contain node attributes, either continuous or discrete attributes? You can transform the discrete attributes into one-hot encodings and concatenate them with continuous attributes as initial node features.

geo47 commented 3 years ago

Yes, I have discrete node attributes. Can you please provide any reference to generate a one-hot encoding representation of node with attributes to generate feature.txt file?

Thanks

ynyunbest commented 3 years ago

Hello, Can you please provide any reference to generate a one-hot encoding representation of node with attributes to generate feature.txt file? Do you have an answer to this question?

Thanks