Figure 1
From: A perturbation-based framework for link prediction via non-negative matrix factorization
Perturbation-based framework by NMF.
Atrain is adjacency matrix of training set, Atest is adjacency matrix of test set, K is the number of latent features, ΔE is perturbation set, A(r) is new perturbed matrix, W(r) is basis matrix, H(r) is coefficients matrix and A* is similarity matrix of original network.
