# What are the methods for constructing GNN input feature vectors?

Which popular methods exist to construct input feature vectors X from the semantic information stored in Knowledge Graph?

In my case, the nodes are 2-3 tokens of text; edges are multi-relation. The objective is to bear semantic similarity (such as “play” and “players” are close) rather than any sort of logical reasoning. Then X would be used to feed R-GNN.

KG:

alice [friend] bob
(football players) [play] (in a team)
bob [play] football


My initial thought is Gensim (and therefore the problem of constructing input feature vectors is out of the scope of GNN), but maybe some other tools/methods exist or provided by DGL or pytorch geometric?