I have a graph of (user, user) and (user, item) relations. How to represent the items?

I have a number of graphs with (user, user) relations and (user, item) relations. Let’s say there are N users and I items. My matrix is R^{NxI} where r_{ni} is how many times user n engaged with item i.

What is the best algorithm for learning an embedding of the items? As I understand, things like metapath2vec would be useful would take into account the weights of the (user, item) relations (ie: How many times a user engaged with an item)

I guess there will be no “best” algorithm. Things ranging from matrix factorization with graph regularization to RGCN can all compute some sort of embeddings for items, but you will need to benchmark on your dataset to pick one that works the best for you.

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.