Improving network embedding with partially available vertex and edge content

Abstract

Network embedding aims to learn a low-dimensional representation for each vertex in a network, which has recently shown its power in many graph mining problems such as vertex classification and link prediction. Most existing methods learn such representations according to network structure information, and some methods further consider vertex content in a network. Unlike prior works, we study the problem of network embedding with two distinctive properties: (1) content information exists on both vertices and edges; (2) only a part of vertices and edges have content information. To solve this problem, we propose a novel Partially available Vertex and Edge Content Boosted network embedding method, namely PVECB, which uses available vertex and edge content information to fine-tune structure-only representations through two hand-designed mechanisms respectively. Empirical results on four real-world datasets demonstrate that our method can effectively boost structure-only representations to capture more accurate proximities between vertices.