Edge Sign Prediction Based on Orthogonal Graph Regularized Nonnegative Matrix Factorization for Transfer Learning

In a social network, the problem often faced is that the network dimension is too high. Nonnegative Matrix Factorization (NMF) is a widely used matrix decomposition method to decrease the high dimensionality. However, the matrix that is generated may not be sparse enough, which can impact its representation ability. To address this problem, we propose Orthogonal Graph Regularized Nonnegative Matrix Factorization (OGNMF) to extract latent features from social networks and prove its convergence theoretically. In scenarios where most of the edge signs are unavailable, conventional learning methods will be ineffective. By comparison, transfer learning methods can improve the learning performance by using another network with adequate signs. After extracting the features, we utilize TrAdaBoost, a classical transfer learning algorithm, to perform experiments. The experimental results using benchmark datasets demonstrate that our method has superior performance to the other baseline methods.