Where to concatenate content features of node? Should it be before attention network?

[kftam1994]

If there are content features of node, such as descriptions and categorical information of the product, where should this frozen embedding (e.g. if sentence embedding with size, number of node*number of embedding dimension) be concatenated in the model?

My understanding is to concatenate to the node embedding before attention network, similar to "Reversed Position Embedding", so it becomes node's representation from graph + Reversed Position Embedding + frozen content features embedding. Is this understanding correct?

Thank you

[CCIIPLab]

Thank you for your attention to our work. We believe that the implementation method you proposed is reasonable. There is also another way, which is to concatenate the context features and node embedding before inputting them into the graph neural network.