Binarized network embedding with community structural information

Network embedding aims to learn low-dimensional vector representations for network nodes by preserving the network structure. The vast majority of existing network embed-ding methods are typically represented in continuous vectors, which impose formidable challenges in storage and computation costs, especially in large-scale applications. To this end, we propose a novel Binarized Network Embedding (BINE) method to learn binary node representations while retaining the community structural information in the network. Specifically, to preserve the community structure among node representations, we absorb the modularity constraint into the matrix factorization-based framework that can approx-imate the first-order and second-order proximity. Meanwhile, the cyclic coordinate des-cent method is adopted to implement the binary node representations for improving storage and computing efficiency. Finally, the conjugate gradient algorithm is employed to jointly optimize the objective function. Extensive experimental results on a variety of real-world network datasets demonstrate that BINE exhibits lower storage and computa-tional complexity than state-of-the-art network embedding methods while achieving com-petitive experimental performance.CO 2022 Published by Elsevier Inc.