Hi, I wonder if there is any explanation on the ACM dataset. I cannot understand what the ‘features’ are after loading the dataset so it is hard for me to apply the model on another dataset. Could you please help on it?
I assume you are referring to the features
here? This should be bag-of-words features for the papers.
Thanks! Do you have any idea what I should do if I would like to use features of various nodes (e.g., paper, author, venue) as input here?
Currently, HAN preprocesses homogeneous graphs over paper out of heterogeneous information. During model computation, message passing/graph convolution is performed and the features of paper nodes get propagated. If you want to employ the features of other types of nodes, you need to construct additional graphs with these kind of nodes.
To set node features of different types you can do something like this:
HG.nodes['node_type1'].data['feature1'] = ...
HG.nodes['node_type2'].data['feature2'] = ...
Hi, In the paper the author mention a technique to project features from different node types into a new feature space by h’ = Wh. However in the implemention I can’t see this part. Could you please help me to figure out how to add this in the code?