Hi, thank you for your great work, it helps me a lot.
Secondly, regarding the loss function in the semi-supervised task, I found that during training, the loss function seems to be different from that in the paper, and only part of it is used. Is my understanding correct, what is the effect of doing this?
In addition, there is another question. I don’t quite understand how to sample two subgraphs. How does the rationale score guide the sampling? Is the final contribution of each node a specific value? Then how to get rationale and non-rationale based on it What about the probability of the rationale part. How to sample after getting the probability?
Hi, thank you for your great work, it helps me a lot.
Secondly, regarding the loss function in the semi-supervised task, I found that during training, the loss function seems to be different from that in the paper, and only part of it is used. Is my understanding correct, what is the effect of doing this?
In addition, there is another question. I don’t quite understand how to sample two subgraphs. How does the rationale score guide the sampling? Is the final contribution of each node a specific value? Then how to get rationale and non-rationale based on it What about the probability of the rationale part. How to sample after getting the probability?