A graph with a single node type & multiple types of edges

[jyoon11]

Hello DGL,

I’m searching for a proper method to solve my link prediction task. To give you a broad context, say, nodes are firms (single type nodes), and edges are of multiple types, such as employee turnover, collaboration, etc.

Given this setting, I’d like to build a link prediction model for predicting future turnover based on not only just the past turnover records (edges) but also other types of network information (e.g., collaboration)

I believe such a network may be seen as a kind of heterogeneous network. Hence, would it be fair to use something like Relational GCN for the link prediction task?

Thanks!

[a651694192]

Yes, I think R-GCN is fair to use here. It has no limit to multiple edges training and the learned node presentation fuse information from all its connected nodes differently according to the edge type, which in my understanding is what you need “not only just the past turnover records (edges) but also other types of network information”.
But original R-GCN don’t use edge attributes for training, not sure if you need it~