Open jiangdie666 opened 1 year ago
Thank you for reaching out. With and without the k-core pruning part will indeed give in different results. Whether the AUC improves or not is actually dataset dependent. The high-level idea is that if the k-core pruning helps to prune more non-informative edges than informative edges, the AUC is expected to increase, and vice versa.
Our main purpose for k-core pruning is to speed up the PaGE-Link algorithm, so it can also run on large graphs. We found that a smaller k will usually do a better job than a larger k of achieving a good balance between accuracy and speed.
I found that synthetic datasets will reveal that the pruning algorithm is boosting the Mask-AUC value when I try the k-core pruning algorithm on the Mask-AUC value. k-core pruning algorithm no k-core pruning algorithm But the k-core pruning algorithm didn't improve the value of Mask-AUC while processing aug_citation, is there any problem in any part. k-core pruning algorithm no k-core pruning algorithm For canceling the k-core pruning algorithm, I do this by changing the prune_graph of the explain function to False.