Gaussian Embedding of Large-Scale Attributed Graphs

Abstract

Graph embedding methods transform high-dimensional and complex graph contents into low-dimensional representations. They are useful for a wide range of graph analysis tasks including link prediction, node classification, recommendation and visualization. Most existing approaches represent graph nodes as point vectors in a low-dimensional embedding space, ignoring the uncertainty present in the real-world graphs. Furthermore, many real-world graphs are large-scale and rich in content (e.g. node attributes). In this work, we propose GLACE, a novel, scalable graph embedding method that preserves both graph structure and node attributes effectively and efficiently in an end-to-end manner. GLACE effectively models uncertainty through Gaussian embeddings, and supports inductive inference of new nodes based on their attributes. In our comprehensive experiments, we evaluate GLACE on real-world graphs, and the results demonstrate that GLACE significantly outperforms stateof-the-art embedding methods on multiple graph analysis tasks.

Deep Multi-Attributed-View Graph Representation Learning

“…We visualize the embedding results of our approach in a two-dimensional space by applying t-SNE [42], as shown in While this work has shown promising performance beyond existing works, graph representation learning still faces great challenges due to real-world networks' complexity. Consequently, one of our future works will be focusing on developing more effective deep graph learning models for dynamic [43], large-scale [44], and heterogeneous graphs [45,46]. Moreover, we tend to explore potential downstream applications of this research, for instance, utilizing multi-attributed-view graph representation learning for social recommendation, anomaly detection, cybersecurity, etc.…”

Robust Attribute and Structure Preserving Graph Embedding

“…Attributed Graph Embedding: Recent studies [1,7,8,11,15,19,21] show that the incorporation of node attributes along with graph structure produces better node embeddings. TADW [19] incorporates text attributes and graph structure with low-rank matrix factorization.…”

“…Noise Modelled Graph Embedding: Most of the existing graph embedding methods represent nodes as point vectors in the embedding space, ignoring the uncertainty of the embeddings. In contrast, Graph2Gauss [1], VGAE [11], DVNE [20] and GLACE [8] capture the uncertainty of graph structure by learning node embeddings as Gaussian distributions. DVNE [20] proposes to measure distributional distance using the Wasserstein metric as it preserves transitivity.…”

“…A great challenge that the existing graph embedding methods face when incorporating both graph structure and node attributes, is the noise prevalent in these two aspects which can mislead the embedding technique to result in learning invalid latent information. Recently, several studies have been proposed to model the uncertainty present in graph data [1,8,11,20]. Most of these work, including VGAE [11], and Graph2Gauss [1], focuses on modelling the uncertainty of the node embeddings by representing the nodes with a probabilistic distribution in the embedding space.…”

$\mathtt{MedGraph:}$ Structural and Temporal Representation Learning of Electronic Medical Records

“…Then, we define joint probability between the two node distribution representations as the likelihood of the existence of a link between them by: P (vi, cj) = Sigmoid(−d(zv i , zc j ))) similarly to the first-order proximity measure in GLACE [15]. Since our graph is unweighted, we can define the prior probability using the structural information observed in the graph as: P (vi,cj) = 1 |Evc| .…”