Multi-level disentanglement graph neural network.
Saved in:
| Title: | Multi-level disentanglement graph neural network. |
|---|---|
| Authors: | Wu, Lirong1,2 (AUTHOR) wulirong@westlake.edu.cn, Lin, Haitao2 (AUTHOR), Xia, Jun2 (AUTHOR), Tan, Cheng2 (AUTHOR), Li, Stan Z.2 (AUTHOR) |
| Source: | Neural Computing & Applications. Jun2022, Vol. 34 Issue 11, p9087-9101. 15p. |
| Subjects: | Deep learning, Diffusion processes |
| Abstract: | Real-world graphs are generally generated from highly entangled latent factors. However, existing deep learning methods for graph-structured data often ignore such entanglement and simply denote the heterogeneous relations between entities as binary edges. In this paper, we propose a novel Multi-level Disentanglement Graph Neural Network (MD-GNN), a unified framework that simultaneously implements edge-level, attribute-level, and node-level disentanglement in an end-to-end manner. MD-GNN takes the original graph structure and node attributes as input and outputs multiple disentangled relation graphs and disentangled node representations. Specifically, MD-GNN first disentangles the original graph structure into multiple relation graphs, each of which corresponds to a latent and disentangled relation among entities. The input node attributes are then propagated in the corresponding relation graph through a multi-hop diffusion mechanism to capture long-range dependencies between entities, and finally the disentangled node representations are obtained through information aggregation and merging. Extensive experiments on synthetic and real-world datasets have shown qualitatively and quantitatively that MD-GNN yields truly encouraging results in terms of disentanglement and also serves well as a general GNN framework for downstream tasks. Code has been made available at: https://github.com/LirongWu/MD-GNN. [ABSTRACT FROM AUTHOR] |
| Copyright of Neural Computing & Applications is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
Be the first to leave a comment!
