You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for bringing such inspiring work! After reading your paper, I understand that how HAN could handle the multiple types of nodes in a graph. But I am not quite sure whether and how HAN could handle the multiple edges.
For example, a user could 'click' and 'purchase' an item. In this case, we have two types of nodes, item, and user; and the nodes are connected via two types of edges, the 'click' and 'purchase'. The natural goal is to derive embeddings for a user under different types of activities. That is, we may want to derive the embedding of the user based on his 'click' behavior, and get another embedding of the user under his 'purchase' activity.
I wonder that whether HAN could handle such tasks. It looks like simply by defining meth-path, say User-Item-User, cannot fully express the types of edges. Any ideas? Thanks in advance!
The text was updated successfully, but these errors were encountered:
Hi! @Jhy1993
Thanks for bringing such inspiring work! After reading your paper, I understand that how HAN could handle the multiple types of nodes in a graph. But I am not quite sure whether and how HAN could handle the multiple edges.
For example, a user could 'click' and 'purchase' an item. In this case, we have two types of nodes, item, and user; and the nodes are connected via two types of edges, the 'click' and 'purchase'. The natural goal is to derive embeddings for a user under different types of activities. That is, we may want to derive the embedding of the user based on his 'click' behavior, and get another embedding of the user under his 'purchase' activity.
I wonder that whether HAN could handle such tasks. It looks like simply by defining meth-path, say User-Item-User, cannot fully express the types of edges. Any ideas? Thanks in advance!
The text was updated successfully, but these errors were encountered: