You mentioned in the article that "we do not treat intra-view nodes as negative samples since the GNN-based encoder is already able to learn intra-graph structure well; we want to focus on inter-graph contrastive information." However, you still used intra loss in the loss function code.
You mentioned in the article that "we do not treat intra-view nodes as negative samples since the GNN-based encoder is already able to learn intra-graph structure well; we want to focus on inter-graph contrastive information." However, you still used intra loss in the loss function code.