DyTed: Disentangled Representation Learning for Discrete-time Dynamic Graph

• Published in: arXiv.org
• Main Authors: Zhang, Kaike, Cao, Qi, Fang, Gaolin, Xu, Bingbing, Zou, Hongjian, Shen, Huawei, Cheng, Xueqi
• Format: Paper Journal Article
• Language: English
• Published: Ithaca Cornell University Library, arXiv.org 16.08.2023
• Subjects: Computer Science - Artificial Intelligence; Computer Science - Learning; Computer Science - Social and Information Networks; Graph representations; Graphical representations; Graphs; Invariance; Invariants; Learning
• ISSN: 2331-8422
• DOI: 10.48550/arxiv.2210.10592

Unsupervised representation learning for dynamic graphs has attracted a lot of research attention in recent years. Compared with static graph, the dynamic graph is a comprehensive embodiment of both the intrinsic stable characteristics of nodes and the time-related dynamic preference. However, existing methods generally mix these two types of information into a single representation space, which may lead to poor explanation, less robustness, and a limited ability when applied to different downstream tasks. To solve the above problems, in this paper, we propose a novel disenTangled representation learning framework for discrete-time Dynamic graphs, namely DyTed. We specially design a temporal-clips contrastive learning task together with a structure contrastive learning to effectively identify the time-invariant and time-varying representations respectively. To further enhance the disentanglement of these two types of representation, we propose a disentanglement-aware discriminator under an adversarial learning framework from the perspective of information theory. Extensive experiments on Tencent and five commonly used public datasets demonstrate that DyTed, as a general framework that can be applied to existing methods, achieves state-of-the-art performance on various downstream tasks, as well as be more robust against noise.