A scalable attribute-aware network embedding system
Network embedding, which aims to generate dense, low-dimensional and representative embedding representations for all nodes in the network, is a crucial step for various AI-based tasks related to network analytics, such as node classification, community detection, and link prediction. In addition to network topology, node attributes are also easily accessible and ubiquitous information that serve as an important role in network embedding. How to jointly incorporate these two kinds of information into unified embedding representations is becoming one of the research focuses on network embedding. However, most existing methods are based on matrix factorization, which is not suitable for relatively large-scale datasets due to their quadratic time and space complexity. Moreover, most real networks evolve over time with varying topological structure or node attributes. This aspect requires the network embedding method to support efficient updating of embedding representations with respect to the evolution of the network. To address these two challenges, we propose a scalable attribute-aware network embedding (SANE) method to efficiently learn the joint embedding representations from topology and attributes. By enforcing the alignment of a locally linear relationship between each node and its K-nearest neighbors in topology and attribute space, the joint embedding representations provide more informative than a single representation extracted from topology or attributes alone. In addition, we devise incremental SANE to support updates of embedding representations in a dynamic environment. Several experiments are conducted on various datasets, demonstrating the effectiveness and scalability of the proposed method.