Constructing Graph Node Embeddings via Discrimination of Similarity Distributions

TL;DR

This paper introduces a novel unsupervised graph node embedding method that discriminates similarity distributions using earth mover distance, achieving state-of-the-art link prediction performance.

Contribution

It proposes a new framework based on discriminating similarity distributions with earth mover distance for unsupervised node embeddings.

Findings

State-of-the-art link prediction results on real-world graphs

Effective discrimination of similarity distributions improves embedding quality

New framework outperforms existing methods in unsupervised learning

Abstract

The problem of unsupervised learning node embeddings in graphs is one of the important directions in modern network science. In this work we propose a novel framework, which is aimed to find embeddings by \textit{discriminating distributions of similarities (DDoS)} between nodes in the graph. The general idea is implemented by maximizing the \textit{earth mover distance} between distributions of decoded similarities of similar and dissimilar nodes. The resulting algorithm generates embeddings which give a state-of-the-art performance in the problem of link prediction in real-world graphs.