Deep Gaussian Embedding of Graphs: Unsupervised Inductive Learning via Ranking

TL;DR

Graph2Gauss introduces a novel unsupervised inductive graph embedding method that models nodes as Gaussian distributions, capturing uncertainty and generalizing to unseen nodes, with strong performance on link prediction and node classification.

Contribution

It is the first to embed nodes as Gaussian distributions for uncertainty modeling and supports inductive learning on various graph types without additional training.

Findings

Outperforms state-of-the-art methods on multiple tasks

Effectively models uncertainty to analyze neighborhood diversity

Generalizes to unseen nodes without retraining

Abstract

Methods that learn representations of nodes in a graph play a critical role in network analysis since they enable many downstream learning tasks. We propose Graph2Gauss - an approach that can efficiently learn versatile node embeddings on large scale (attributed) graphs that show strong performance on tasks such as link prediction and node classification. Unlike most approaches that represent nodes as point vectors in a low-dimensional continuous space, we embed each node as a Gaussian distribution, allowing us to capture uncertainty about the representation. Furthermore, we propose an unsupervised method that handles inductive learning scenarios and is applicable to different types of graphs: plain/attributed, directed/undirected. By leveraging both the network structure and the associated node attributes, we are able to generalize to unseen nodes without additional training. To learn the embeddings we adopt a personalized ranking formulation w.r.t. the node distances that exploits the natural ordering of the nodes imposed by the network structure. Experiments on real world networks demonstrate the high performance of our approach, outperforming state-of-the-art network embedding methods on several different tasks. Additionally, we demonstrate the benefits of modeling uncertainty - by analyzing it we can estimate neighborhood diversity and detect the intrinsic latent dimensionality of a graph.