Deep Attributed Network Embedding with Community Information

Attributed Network Embedding (ANE) aims to learn low-dimensional representation for each node while preserving topological information and node attributes. ANE has attracted increasing attention due to its great value in network analysis such as node classification, link prediction, and node clustering. However, most existing ANE methods only focus on preserving attribute information and local structure, while ignoring the community information. Community information reveals an implicit relationship between vertices from a global view, which can be a supplement to local information and help improve the quality of embedding. So, those methods just produce sub-optimal results for failing to preserve community information. To address this issue, we propose a novel method named DNEC to exploit local structural information, node attributes, and community information simultaneously. A novel deep neural network is designed to preserve both local structure and node attributes. At the same time, we propose a community random walk method and incorporate triplet-loss to preserve the community information. We conduct extensive experiments on multiple real-world networks. The experimental results show the effectiveness of our proposed method.