Hetero GNN Link prediction #4643
-
|
problem statement: I m new to pyg and I need a recommendation system using a graph neural network. We have to recommend skills to users based on their current performance. For an instant, if a user had already taken some skills and is current if we is taking X skill and their proficiency with X skill is 70 percent (or perform well with the skill x ) then he/she should be shown the next post skill of that X. if he/she does not perform well than recommend the previous skill of X. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 11 replies
-
|
This looks like a heterogeneous link prediction problem to me. You can find an example for this here. |
Beta Was this translation helpful? Give feedback.
-
|
Adding reverse relations lets the message passing phase pass messages in both directions. Otherwise, only destination node types will get updated, while source nodes stay with their original representation. It is a common strategy to boost performance in directional graphs (using distinct parameters for each direction). |
Beta Was this translation helpful? Give feedback.
-
|
@rusty1s I m having this error while splitting the train, test, and validation edges |
Beta Was this translation helpful? Give feedback.
-
|
@rusty1s I have changed the way of data passing to edge_type, re_edge_type parameter. Now its giving me this issue. |
Beta Was this translation helpful? Give feedback.
-
|
I think this is already fixed in master. Can you try to follow these instructions and test again? Sorry for the inconvenience! |
Beta Was this translation helpful? Give feedback.
-
|
@rusty1s thanks it worked with the master. Now I have this issue |
Beta Was this translation helpful? Give feedback.
This looks like a heterogeneous link prediction problem to me. You can find an example for this here.