PROJECT TITLE :

Large Scale Network Embedding A Separable Approach

ABSTRACT:

There have been many successful methods proposed for learning low-dimensional representations on large-scale networks; however, almost all of the methods that are currently in use are designed in inseparable processes, meaning that they learn embeddings for entire networks even though only a small proportion of nodes are of interest. This results in a significant amount of inconvenience, particularly on large-scale or dynamic networks, where it becomes difficult, if not impossible, to implement these methods. In this paper, we formalize the problem of separated matrix factorization. On the basis of this formalization, we elaborate a novel objective function that preserves both local and global information. We compare our SMF framework with approximate SVD algorithms and demonstrate that SMF can factorize a given matrix while capturing more information than the other method. In addition, we propose an algorithm for network embedding called SepNE. It is uncomplicated and adaptable, and it can independently learn representations for various subsets of nodes using separate learning processes. Our algorithm achieves scalability to large networks by reducing the redundant efforts to embed irrelevant nodes. This is accomplished through the implementation of separability. To further incorporate complex information into SepNE, we discuss several methods that can be used to leverage high-order proximities in large networks. This will allow us to incorporate additional complex information. We demonstrate the effectiveness of SepNE on a variety of real-world networks ranging in scale and covering a variety of topics. Our method significantly outperforms the state-of-the-art baselines in terms of running times on large networks, despite having accuracy that is comparable to theirs.