question about the code

[KYRA-ma]

Hi, I want to express my admiration for your idea first of all.

However, I have some questions about the code. When you obtain the latent feature matrix $\mathbf{Y}{i}$，it should be solve by the eigenvectorscorresponding to the first $d$ smallest eigenvalues of $\mathbf{L}{i}$. But in your code, it seems to get the first $d$ largest eigenvalues of $\mathbf{L}_{i}$.

Or maybe I misunderstood. Hope to get your answer, thx.@WangJun2023

[WangJun2023]

Thanks for your interest in our work. The standard Laplacian matrix is L = I - D^-1/2WD^-1/2, in our Matlab code, L is denoted by D^-1/2WD^-1/2， and the subtraction process is omitted. So we get the first d largest eigenvalues in the code.

[KYRA-ma]

i got it. Thx.

[KYRA-ma]

I'm sorry that I'm still confused about the SC algorithm for generating the groups in the code file 'clu_ncut.m'.

I'm confused that why we perform singular value decomposition on the standard Laplacian matrix but not eigenvalue decomposition here according to the SC theorm？And why here get the first k largest eigenvectors in matrix U but not the fist k smallest one or matrix V?

Hope to get your answer again, thanks!