search

divelab / DIG

A library for graph deep learning research
https://diveintographs.readthedocs.io/
GNU General Public License v3.0
1.84k stars 281 forks source link

GNNExplainer: optimization on the whole graph #110

Closed alirezadizaji closed 2 years ago

alirezadizaji commented 2 years ago

Hi as I understood, the only option for node-based explanation is to choose a subgraph with the same k-hop with the original model. is it possible to choose the k-hop of GNNExplainer arbitrarily? I believe the subset generated by the subgraph method could help this matter.

Oceanusity commented 2 years ago

Hello, since the node information aggregates from its k-hop neighbors for a k-layer GNN, it's defaulted to explain the importance of its k-hop neighbors.

Oceanusity commented 2 years ago

Such k-hop neighbors all contribute to the final prediction of this node. Therefore, the explainability method such as GNNExplainer should consider them all in the algorithm.