TG-Gen: A Deep Generative Model Framework for Temporal Graphs

Ryien Hosseini, Filippo Simini

Keywords: graph neural networks, generative models, temporal graphs

Abstract: Graph Neural Networks (GNNs) have recently emerged as popular methods for learning representations of non-euclidean data often encountered in diverse areas ranging from chemistry and biology to social and financial networks. More recently, research has focused specifically on learning on temporal graphs, wherein the nodes and edges of a graph, and their respective features, may change over time. However, existing work in the temporal graph space has largely focused on discriminative models. In this work, we present TG-Gen, a generic generative framework for temporal graph data, which combines an encoder module that creates temporal embeddings of nodes from raw interaction data, with a decoder module that uses the learned temporal embeddings to create a deep probabilistic model of interaction data. We show that TG-Gen is able to generate robust and accurate synthetic data for temporal graphs for two traditional benchmark data and a novel dataset. Additionally, we demonstrate that TG-Gen is able to learn generalizable representations of temporal graphs and outperforms the previous state-of-the-art method in the discriminative regime, such as for dynamic link prediction. Finally, we perform comprehensive ablation studies which show the effects of specific modules and configurations of our model.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Generative models

TL;DR: We propose, TG-Gen, a generic framework for generating synthetic temporal graph data.