Attributed Network Embedding Using an Improved Weisfeiler-Lehman Schema and a Novel Deep Skip-Gram

Abstract

Attributed Network Embedding (ANE) and the representation of its nodes in a low-dimensional space is a pivotal step in the analysis of real-world networks. One of the biggest challenges in the embedding process of nodes in complex networks is to capture any dynamic changes in both the node itself and in its adjacent. To address the above challenge, in this paper, we propose a novel ANE model that combines an improved Weisfeiler-Lehman Information Aggregation (WLIA) schema with a novel Deep Skip-Gram (DSG) approach. First, an information aggregation of network data is performed using an improved Weisfeiler-Lehman, which captures each node’s attributes and combines them with the attributes of its adjacent nodes in a mathematically proven balanced and fair manner. Next, a novel deep autoencoder model that adopts the Skip-Gram approach to capture the high non-linearity among the nodes and between nodes with their attributes is proposed. In the DSG approach, a deep encoder is paired with a set of deep decoders; the main decoder is for the node itself and the secondary deep decoders act as attention decoders to extract common features from its neighbors. Extensive experimental evaluations have demonstrated that the proposed method is superior in performance compared to recent network embedding models.

Keywords

• Attributed network embedding
• Weisfeiler-Lehman
• skip-gram
• random walks
• autoencoder