Dual Graph Representation Learning

Huiling Zhu, Xin Luo, Hankz Hankui Zhuo

Abstract: Graph representation learning embeds nodes in large graphs as low-dimensional vectors and benefit to many downstream applications. Most embedding frameworks, however, are inherently transductive and unable to generalize to unseen nodes or learn representations across different graphs. Inductive approaches, such as GraphSAGE, neglect different contexts of nodes and cannot learn node embeddings dually. In this paper, we present an unsupervised dual encoding framework, \textbf{CADE}, to generate context-aware representation of nodes by combining real-time neighborhood structure with neighbor-attentioned representation, and preserving extra memory of known nodes. Experimently, we exhibit that our approach is effective by comparing to state-of-the-art methods.