Open ChuanjiuWu opened 2 years ago
https://arxiv.org/pdf/2002.02126.pdf The model in original paper didn't use Laplace matrix. I think you can also use Laplace matrix, but this two forms might have different performances for different tasks.
Thank you for your reply. I may have misrepresented, what I was asking is where to reflect the symmetrically normalized matrix
Why didn't Lightgcn's convolutional layer use the Laplace matrix? If this is not the case, can you please point out which line of code it is reflected in....